See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/259535572

Personal Information Management: From Information Consumption to Curation

Article  in  Annual Review of Information Science and Technology · January 2011

DOI: 10.1002/aris.2011.1440450108

CITATIONS READS
143 1,129

1 author:

Steve Whittaker
University of California, Santa Cruz
320 PUBLICATIONS   14,744 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

A lifespan study of the developmental affordances of solitude View project

All content following this page was uploaded by Steve Whittaker on 27 April 2020.

The user has requested enhancement of the downloaded file.

 1  

Personal  Information  Management:  From  Information  Consumption  to  Curation  

Steve  Whittaker  

IBM  Research  Almaden  

ABSTRACT    
An   implicit,   but   pervasive   view   in   the   information   science   community   is   that   people   are  
perpetual   seekers   after   new   public   information,   incessantly   identifying   and   consuming   new   information  
by   browsing   the   web,   and   accessing   public   collections.   One   aim   of   this   review   is   to   challenge   this  
consumer  characterisation  which  regards  information  as  a  public  resource  containing  novel  data  that  we  
seek   out,   consume,   and   then   discard.   Instead   I   want   to   focus   on   a   very   different   view:   where   familiar  
information  is  used  a  personal  resource  that  we  keep,  manage  and  (sometimes  repeatedly)  exploit.  I  call  
this   information   curation.   I   first   summarise   arguments   against   the   consumer   perspective.   I   then   review  
research   on   three   different   information   curation   processes:   keeping,   management   and   exploitation.   I  
describe  existing  work  detailing  how  each  of  these  processes  is  applied  to  different  types  of  personal  data,  
namely  documents,  emails,  photos  and  web  pages.  The  research  indicates  people  tend  to  keep  too  much  
information   with   the   exception   of   contacts   and   web   pages.   When   managing   information   there   are  
surprising   benefits   for   strategies   that   rely   on   piles   as   opposed   to   files.   And   despite   the   emergence   of  
desktop  search,  exploitation  currently  remains  reliant  on  manual  methods  such  as  navigation.  There  are  a  
number   of   new   technologies   that   could   potentially   address   important   curation   problems,   but  
implementing   these   in   acceptable   ways   remains   a   challenge.   I  conclude  with   a   summary   of   outstanding  
research  and  technical  questions.  

1.INFORMATION  SEEKING  AND  CONSUMPTION  

There   is   a   long   tradition   within   information   and   computer   science   of   defining   information   in  
terms   of   its   novelty   and   its   ability   to   transform   whoever   consumes   it   (Shannon   and   Weaver,   1949).  
Consistent  with  this,  a  new  set  of  computer  science  theories  argue  that  our  general  information  behaviour  
is  akin  to  the  foraging  behaviours  of  hunter  gatherer  peoples  (Pirolli,  2007,  Pirolli  and  Card,  1995,  1999).  
According   to   this   view,   people   actively   seek   out   and   consume   new   information   from   public   resources.  
While   not   as   extreme   in   its   focus   on   consumption,   the   information   science   literature   also   emphasises  
discovery   of   new   public   information,   rather   than   its   exploitation.   Many   different   models   have   been  
proposed   to   characterise   how   people   find   new   information   in   public   collections   (Belkin,   1980,   Ellis   and  
Haugan,   1997,   Kuhlthau,   1991,   Marchionini,   1995,   Wilson,   1981,   1994).   As   with   information   foraging,  
these   models   focus   exclusively   on   the   process   of   locating   new   information   from   public   resources   (e.g.  
archives   or   the   web).   Although   such   information   seeking   is   acknowledged   to   be   iterative,   with   people  
making  repeated  short-­‐term  efforts  to  satisfy  information  needs  (Belkin,  1980,  Marchionini,  1995),  these  
models  are  silent  about  what  happens  once  such  valuable  information  is  located.  They  do  not  discuss  how  
this  information  is  organised  or  curated  for  future  use.    

For   example,   in   a   very   influential   theory,   Belkin   (1980)   proposes   that   people   are   motivated   by  
ASK   (an   anomalous   state   of   knowledge)   to   discover   new   relevant   information.   He   talks   about   the   steps  
 2  

that   people   follow   to   address   that   anomalous   state.   Kuhlthau   (1991)   describes   various   information  
seeking  processes,  including  recognising  an  information  need,  and  identifying  a  general  topic,  as  well  as  
stages   for   formulating   and   gathering   information.   A   similar   feature   set   model   is   proposed   by   Ellis   and  
Haugan   (1997)   who   detail   the   activities   involved   in   finding   information,   including   browsing,   chaining,  
monitoring,   differentiating,   extracting   and   verifying   information.   Wilson   (1999)   provides   a   high   level  
macro-­‐model   which   characterises   how   information   needs   arise,   and   what   aids   or   hinders   these   processes  
of   information   seeking,   incorporating   insights   from   these   lower   level   accounts.   Marchionini’s   (1995)  
model  is  focused  on  more  recent  technologies,  discussing  how  information  seeking  moves  from  high  level  
framing   of   information   needs   to   expressing   those   as   some   form   of   query,   evaluation   of   the   results   of  
executing  that  query  and  reiteration  depending  on  the  outcome  of  that  evaluation.    

However,   all   these   models   talk   only   about   how   public   information   is   found,   and   ignore   what  
happens   after   finding   has   occurred.   In   a   systematic   meta-­‐analysis   of   theoretical   information   science,  
Wilson  (1999)  confirms  that  information  science  theories  have  not  tackled  what  he  calls  ‘information  use’,  
i.e.  what  happens  after  information  seeking  is  completed.  I  will  argue  that  this  emphasis  on  information  
seeking   is   based   on   a   partial   and   unrepresentative   view   of   what   people   usually   do   with   information.   In  
contrast  to  the  foraging  and  information  seeking  viewpoints,  this  review  is  concerned  with  an  increasingly  
important   (but   very   different)   set   of   behaviours   which   I   call   personal   information   curation.   While   there  
are   some   studies   of   curation   behaviours   in   the   collection   management   literature,   these   have   tended   to  
focus   on   the   activities   of   information   professionals   who   are   trained   to   organise   and   manage   public  
collections   (Drew   and   Dewe,   1992,   Osburn   and   Atkinson,   1991).   Similar   studies   within   computer  
supported   co-­‐operative   work   look   at   how   teams   self-­‐organise   to   create   shared   repositories   (Ackerman  
1998,   Ackerman   and   Halverson,   2004,   Berlin   et   al.,   1993).   In   both   these   cases,   however,   the   focus   of  
curation  is  on  organisation  of  public  and  not  personal  collections.  Here  I  will  review  evidence  showing  that  
people’s   everyday   information   habits   are   frequently   focused   around   managing   personal   data   and   don’t  
involve  incessant  access  and  immediate  consumption  of  new  public  information.  Instead  people  keep  and  
manage  personal  information  for  future  exploitation.  While  reviewing  the  general  literature,  I  will  provide  
illustrative  examples  of  each  of  these  behaviours  from  my  own  research  and  that  of  my  collaborators.    

1.1  CURATION  IS  THE  RULE  AND  NOT  THE  EXCEPTION  

One  very  strong  argument   for   the   incompleteness   of   the   consumption   model   is   that   people   keep  
personal   information.   Information   seeking   and   foraging   models   argue   that   we   are   continually   seeking   out  
novel  public  resources.  If  these  models  are  correct,  then  we  should  not  expect  people  to  conserve  large  
amounts   of   information   for   future   consumption.   However,   a   minute’s   reflection   will   reveal   that   that  
people   persistently   engage   in   active   and   extensive   preservation   and   curation   behaviours   in   their  
information   environments.   Much   as   we   might   want   to,   we   don’t   immediately   delete   each   email   we  
receive,   once   we   have   read   or   replied   to   it.   And   after   creating   a   document   or   presentation,   we   don’t  
immediately  transfer  it  to  the  trash.  We  take  care  to  preserve  personal  photos  over  periods  of  years.    

There   are   many,   many,   examples   of   people   preserving   and   managing   personal   materials   for  
future  exploitation.  Here  are  some  simple  statistics  about  the  huge  amounts  of  information  that  people  
keep  in  their  personal  stores.  Whittaker  et  al.,  (2007)  summarise  8  studies  of  email,  showing  that  people  
archive   a   huge   number   of   messages,   with   an   average   of   around   2846   messages   being   kept.  
Unsurprisingly,  these  personal  email  archives  are  growing  larger,  with  more  recent  studies  (Fisher  et  al.,  
2006)  revealing  that  people  have  around  28,000  messages.  People  also  keep  a  large  number  of  personal  
 3  

files.   Boardman   and   Sasse   (2004)   found   an   average   of   around   2200   personal   files   stored   on   people’s   hard  
drives.  And  a  recent  study  of  digital  photos  found  an  average  of  over  4000  personal  pictures  (Whittaker  et  
al.,  2010).  Studies  of  web  bookmarking  show  that  people  also  preserve  hundreds  of  bookmarks  (Abrams  
et  al.,  1998,  Aula  et  al.,  2005,  Boardman  and  Sasse,  2004,  Cockburn  and  Greenberg,  2000,  Catledge  and  
Pitkow,   1995).   And   of   course   these   behaviours   aren’t   limited   to   the   digital   domain:   Whittaker   and  
Hirschberg  (2001)  looked  at  paper  archives  and  found  that  people  still  amassed  huge  amounts  of  personal  
paper  data.  That  study  found  that  on  average  researchers  had  62kg  of  paper,  equivalent  to  a  pile  of  phone  
directories  30m  high.    

Furthermore  it’s  not  just  that  people  passively  keep  this  information,  they  also  make  strenuous  
attempts   to   organise   it   in   ways   that   will   promote   future   retrieval.   For   email,   Bellotti   et   al   (2005)   found  
that   people   spend   10%   of   their   total   time   in   email   filing   messages,   leading   to   an   average   of   244   folders   in  
their  email  collections.  Personal  computer  files  are  organised  in  a  similar  way,  with  people  averaging  57  
folders  with  an  average  depth  of  3.3  subfolders  (Boardman  and  Sasse,  2004).  Studies  of  web  bookmarking  
also   show   active   organisational   efforts   leading   to   an   average   of   17   folders   with   complex   subfolder  
structure   (Abrams   et   al.,   1998,   Aula   et   al.,   2005).   And   Marshall   (2008a,b)   describes   the   arcane  
organisations  that  result  from  attempts  to  preserve  information  over  many  years.    

So  while  it’s  obvious  that  consumption  is  important  for  some  types  of  rapidly  changing  transient  
public   information   (news,   entertainment),   it   is   not   the   norm.   For   most   types   of   information   behaviour  
seems  to  be  much  closer  to  curation  than  consumption.  Furthermore,  curation  seems  destined  to  become  
even   more   important.   New   technologies   such   as   ubiquitous   sensors,   digital   video   and   digital   cameras,  
make   it   increasingly   easy   to   capture   new   types   of   personal   data.   And   this   trend,   along   with   continued  
increases   in   cheap   digital   storage   mean   that   people’s   hard   drives   are   now   filling   up   with   huge   amounts   of  
personal  photos,  videos  and  music  (Bell  and  Gemmell,  2009,  Kalnikaite  et  al.,  2010,  Marshall,  2008a,b).  

One   obvious   objection   to   the   argument   for   curation   is   that   we   spend   large   amounts   of   time  
accessing   public   resources   such   as   the   web.   However   new   research   shows   that   even   here   we   aren’t  
seeking   out   novel   information.   Accessing   public   the   web   usually   entails   reaccessing   previously   visited  
resources.   Various   studies   have   shown   that   most   of   people’s   web   behaviour   concerns   reaccess,   i.e.  
returning  to  information  they  have  already  looked  at.  Between  58  and  81%  of  all  user  web  accesses  are  of  
pages   that   the   user   has   accessed   previously   (Cockburn   and   Greenberg,   2000,   Obendorf   et   al.,   2007,  
Tauscher  and  Greenberg,  1997).  So,  rather  than  people  foraging  for  new  information  and  resources,  they  
instead  revisit  previously  accessed  information.  Again  this  suggests  a  pattern  of  curation  and  re-­‐use  rather  
than  one-­‐time  consumption.  

If  these  arguments  are  correct,  we  need  to  rethink  our  theories  of  information.  Prior  systems  and  
models   of   information   describe   consumption   of   public   data.   Indeed   until   recently   it   wasn’t   possible   to  
create   and   keep   significant   personal   digital   archives.   However   the   prevalence   of   keeping   and   re-­‐use  
suggest   a   need   to   develop   theories   of   curation,   i.e.   the   active   preservation   of   personal   information  
content   for   the   future.   We   need   to   look   beyond   models   of   Foraging   and   Information   Seeking   to   think  
about   practices   of   Preserving   and   Curating   information.   Agricultural   practices   allowed   our   ancestors   to  
free   themselves   from   the   vagaries   of   an   unpredictable   environment.   In   the   same   way,   we   need   new  
theories,   tools   and   practices   for   Information   Curation   to   help   support   these   pervasive   activities.   While  
other  work  has  neglected  how  we  acquire  and  manage  personal  information,  one  exception  is  the  work  of  
 4  

Jones  and  colleagues  (Bruce  et  al.,  2004,  Jones,  2004,  2007,  Jones  and  Teevan,  2007),  and  we  use  a  variant  
of  Jones’  PIM  lifecycle  framework  to  organise  this  review.    

The  structure  of  the  chapter  is  as  follows.  In  section  2  we  present  a  framework  for  the  curation  
lifecycle,  which  describes  the  processes  by  which  we  Keep,  Manage  and  Access  information,  elaborating  
the   relationships   between   these   processes.   We   also   discuss   important   distinctions   between   different  
properties  of  information  that  have  implications  for  curation,  such  as  whether  information  is  unique  and  
whether  it  requires  action.  The  next  three  sections  then  review  the  challenges  of  Keeping,  Management  
and   Exploiting   personal   information.   We   present   relevant   research   on   how   and   why   people   keep  
information,   the   different   ways   they   organise   it,   and   finally   how   they   access   and   exploit   that   stored  
information.   In   each   case   we   review   how   different   types   of   information   (emails,   documents,   photos,  
webpages)   are   treated   differently.   The   final   section   looks   to   the   future,   exploring   different   technical  
developments   that   may   influence   the   future   of   information   curation,   as   well   as   outlining   outstanding  
empirical  and  methodological  issues.  

2.THE  CURATION  LIFECYCLE    

Curation   involves   future   oriented   activities,   more   specifically   the   set   of   practices   that   select,  
maintain   and   manage   information   in   ways   that   are   intended   to   promote   future   consumption   of   that  
information.  We  begin  by  introducing  a  simple  3  stage  model  of  the  curation  lifecycle  that  is  a  variant  of  
that  described  in   (Jones,  2007,  Jones  and  Teevan,  2007).  We  talk  about  the  relations  between  different  
phases   of   the   lifecycle,   and   clarify   differences   between   our   framework   and   Jones’   work.   We   also  
introduce   important   distinctions   between   different   properties   of   information   that   have   implications   for  
curation  behaviours.    

2.1  KEEPING    
We   encounter   new   information   all   the   time.   Much   of   this   encountered   information   may   be  
irrelevant  to  us,  and  other  pieces  of  information  such  as  news  or  trivia  are  of  little  future  utility  once  we  
have  registered  them.  But   some  of  this  new  information  we  anticipate  needing  in  the  future.  But  how  do  
we   decide   what’s   worth   keeping?   What   principles   govern   decisions   about   the   sorts   of   information   we  
keep  (Jones,  2004,  2007)?  There  are  costs  to  keeping,  so  how  do  we  decide  which  information  will  have  
significant   future   value,   and   what   makes   it   worth   keeping   (Marshall,   2008a,b)?   Keeping   is   clearly   a  
complex   decision   that   is   influenced   by   many   factors,   including   the   type   of   information   being   evaluated,  
when  we  anticipate  will  need  it,  as  well  as  the  context  in  which  we  imagine  that  it  will  be  needed.  There  
are   also   strategic   trade-­‐offs   involved   in   keeping   information   ourselves   rather   than   relying   on  regenerating  
that  same  information  from  public  resources.  

Information   items   (whether   they   are   documents,   emails,   photos   or   webpages)   have   different  
utility   and   will   consequently   be   processed   in   very   different   ways.   Transient   information   encountered   on   a  
web   page   will   be   treated   very   differently   from   a   personal   document   we   have   been   working   on   for   several  
days,   or   an   email   sent   by   an   important   colleague.   The   technologies   that   we   use   to   generate   and  
encounter   information   will   also   have   an   effect   on   how   likely   we   are   to   keep   it.   For   example   digital  
photography  has  now  made  it  much  easier  to  take  very  many  pictures.  And  preserving  digital  pictures  is  
inexpensive   because   storage   technology   is   now   so   cheap.   One   consequence   is   that   people   are   keeping  
many  more  pictures,  compared  with  the  past  when  taking  pictures  was  expensive,  developing  them  was  
 5  

laborious  and  they  required  careful  physical  organisation  and  storage.  But  the  ease  of  generating  pictures  
may   have   important   downstream   consequences   for   retrieval   that   need   to   be   taken   into   account   when  
deciding  whether  or  not  to  keep  them  (Whittaker  et  al.,  2010).  

2.2  MANAGEMENT  
Having   decided   that   we   want   to   keep   certain   information,   how   should   we   manage   that  
information   in   ways   that   will   guarantee   it   will   produce   future   value?   Again   this   depends   on   the   type   of  
information,  and  again  there  are  strategic  questions.  A  key  decision  people  have  to  make  is  the  trade-­‐off  
between  the  effort  to  invest  in  managing  information,  against  the  projected  payoff  during  exploitation.    

There   are   different   ways   of   managing   information   that   have   different   costs   and   payoffs.   As  
information   curators,   we   have   to   decide   between   intensive   methods   that   are   likely   to   engender   higher  
information   yields   but   at   the   cost   of   greater   management   efforts.   These   intensive   methods   must   be  
compared  with  less  intensive  methods  that  may  guarantee  less  predictable  returns.  For  example  we  might  
apply   systematic   structure   to   our   paper   files,   e.g.   by   filing   our   incoming   information   into   structured  
folders.   This   information   should   then   be   more   straightforward   to   access   -­‐   providing   that   the   structures  
match   the   context   in   which   we   wish   to   retrieve   the   information.   However   this   filing   strategy   imposes   a  
heavy  burden  on  the  information  curator  because  each  new  piece  of  information  must  be  analysed  and  
structured   in   this   way.   Alternatively   we   may   adopt   a   more   relaxed   approach   and   allow   physical  
information  to  accumulate  in  piles  on  our  desk,  or  emails  to  pile  up  in  our  inbox.  This  tactic  reduces  the  
costs  of  organising  this  information,  but  may  mean  that  it  is  harder  to  locate  critical  information  when  we  
need  it  (Malone,  1983,  Whittaker  and  Hirschberg,  2001,  Whittaker  and  Sidner,  1996,  Whittaker,  2005).  

The  management  process  is  also  organic  and  we  modify  our  personal  information  systems  in  an  
adaptive   way.   We   repeatedly   revisit   and   restructure   information   related   to   ongoing   tasks   to   meet   our  
current  needs.  People  may  be  able  to  remember  more  about  the  organisation  of  recently  or  frequently  
visited  information  -­‐  making  it  straightforward  to  access.  Other  types  of  information  may  be  infrequently  
accessed   –   e.g.   photos   that   are   stored   for   the   long   term.   Infrequent   access   may   mean   that   users   don’t  
discover   that   their   photo   collection   needs   to   be   systematically   restructured   for   it   to   be   effectively  
retrieved  (Whittaker  et  al.,  2010).      

Management   may   also   have   repetitive   properties.   Some   people   habitually   ‘weed   out’  
information   that   has   turned   out   to   be   of   little   value   that   may   be   compromising   the   uptake   of   information  
of  definite  utility.  People  occasionally  work  through  email  inboxes  deleting  old  or  irrelevant  information  
(Whittaker   and   Sidner,   1996).   However   it   is   abundantly   clear   that   people   find   such   ‘cleanup’   activities  
difficult,   not   only   because   they   require   judgements   about   the   projected   value   of   information,   but   also  
there  may  be  emotional  investment  in  information  that  they  have  invested  time  and  effort  in  organising  
(Bergman  et  al.,  2003,  Jones,  2007,  Marshall,  2008a,  b,  Whittaker  and  Hirschberg,  2001).  

2.3  EXPLOITATION  
Exploitation  is  at  the  heart  of  curation  practices.  If  we  cannot  successfully  exploit  the  information  
we   preserved,   then   keeping   decisions   and   management   activity   will   have   been   futile.   But   what   are  
effective   ways   for   accessing   curated   information?   Exploitation   success   clearly   relates   to   keeping   and  
management   practices.   Careful   attempts   to   organise   valuable   information   should   make   it   easier   to  
reaccess  that  data.  But  new  technologies  potentially  reduce  the  need  to  organise.  Emerging  technologies  
 6  

such  as  desktop  search  (Cutrell  et  al.,  2006a,  Dumais  et  al.,  2003,  Russell  and  Lawrence,  2007),  promise  to  
reduce  the  overhead  of  organising  our  files,  because  we  no  longer  have  to  manually  navigate  to  them.    

There  are  two  main  methods  that  can  be  used  to  exploit  information  

- Navigation  –  which  exploits  structures  the  user  has  set  up  for  retrieval  and  involves  incremental  
manual  traversal  of  these  structures.  
- Search  –  a  more  indirect  way  to  find  information  –  where  the  user  generates  textual  labels  that  
refer  to  the  name  of  information  item,  one  of  its  attributes  or  its  contents.    
 
There  are  advantages  and  disadvantages  of  both  methods.  Navigation,  being  incremental,  offers  
the  user  feedback  at  each  access  stage  (Barreau  and  Nardi,  1995,  Bergman  et  al.,  2008),  but  in  the  case  of  
complex  folder  hierarchies  can  be  laborious  because  of  the  multiple  levels  that  people  have  to  traverse.  
Search  is  potentially  more  flexible  allowing  users  to  specify  multiple  properties  of  the  target  file  (Lansdale,  
1988).  However  it  is  reliant  on  being  able  to  remember  salient  properties  of  the  target  item  in  order  to  
generate  appropriate  search  terms.    

Relation  to  Jones’  PIM  framework  

The  differences  in  terminology  between  our  framework  and  that  of  Jones  and  Teevan  (2007)  and  
Jones  (2007)  are  shown  in  Table  1.  The  frameworks  concur  in  their  overall  characterisation  of  key  personal  
information   management   processes,   such   as   keeping   and   management.   However   Jones   and   Teevan  
(2007)’s  main  focus  is  on  finding  and  refinding  of  public  data  such  as  that  found  on  the  web,  e.g.  if  people  
want   to   repeatedly   access   a   valued   web   resource.   In   contrast,   in   this   review   we   are   more   concerned   with  
information  that  people  create  themselves  or  that  they  receive  in  email.  Where  we  focus  on  web  data  it  is  
in   the   context   of   users’   efforts   to   integrate   such   information   into   their   personal   information   systems.   Our  
more   strict   definition   of   personal   information   means   we   begin   our   model   with   keeping.   Keeping   is   a  
prerequisite   for   later   stages:   people   cannot   manage   or   exploit   information   that   they   have   not   kept.   In  
contrast  because  Jones  and  Teevan  (2007)’s  concern  is   more   with   public   data,  they  begin  with  (re)finding  
such   information,   because   it   already   exists   in   the   public   domain  without   users   making   efforts   to   create   or  
preserve  it.  

PIM ACTIVITIES CURATION LIFECYCLE

JONES, 2007, JONES AND TEEVAN, 2007

(RE)FINDING
KEEPING KEEPING
METALEVEL ACTIVITIES (MANAGING , MANAGEMENT
MAINTAINING , ...)
EXPLOITATION
     

Table  1:  Contrast  between  PIM  activities  proposed  by  Jones  and  Teevan  (2007)  and  those  used  in  
the  current  review  

2.4  INTERRELATIONS  BETWEEN  KEEPING,  MANAGEMENT  AND  EXPLOITATION  

 7  

As   will   be   obvious   from   the   above   description,   there   are   close   relations   between   the   different  
processes   in   the   curation   lifecycle.   For   example,   exploitation   success   is   highly   dependent   on   the  
information  people  choose  to  preserve,  as  well  as  the  method  they  use  to  manage  it.  Keeping  information  
does  not  necessarily  guarantee  that  it  will  be  successfully  exploited,  and  the  more  information  we  keep,  
the  more  effort  has  to  go  into  organising  and  maintaining  that  information.  More  critically,  having  more  
information  may  increase  the  difficulty  of  exploitation,  as  finding  information  may  be  harder  when  there  
is  more  information  to  search.    

Past   outcomes   may   also   influence   future   curation   behaviours.   Past   exploitation   success   may  
influence   future   keeping   and   management   practices.   If   certain   information   is   difficult   to   re-­‐access   or  
maintain,   people   may   conclude   that   there   is   little   point   in   keeping   it   in   future.   In   the   same   way,  
exploitation  failure  may  cause  people  to  change  their  management  methods.  If  users  realise  that  certain  
types  of  organisation  are  less  successful  in  promoting  access  they  may  abandon  those  methods.    

2.5  INFORMATION  PROPERTIES  

Not  all  information  items  are  equivalent.  We  need  to  distinguish  between  different  information  
properties,  as  these  differences  have  implications  for  the  ways  in  which  each  type  of  item  will  be  curated.  

Informative  versus  Action-­‐oriented  items    

Compare,   for   example,   an   average   email   message,   with   a   page   found   in   a   web   search.   One  
crucial  property  of  many  email  messages  is  that  they  require  the  recipient  to  do  something,  whether  it  is  
to   respond   to   a   question,   arrange   a   meeting,   or   provide   some   information.   Such   emails   are   action  
oriented,   because   the   message   recipient   is   expected   to   respond   in   some   way,   often   within   a   specific  
timeframe   (‘let   me   know   about   this   within   the   next   day’).   In   contrast   information   items   found   during   a  
web  search  are  potentially  informative  -­‐  but  do  not  usually  require  users  to  act.  While  the  page  may  be  
diverting,   there   is   no   requirement   to   process   the   information   on   the   page   to   meet   a   given   deadline.   Of  
course   this   Information   vs.   Action   distinction   does   not   map   neatly   to   computer   applications.   Not   every  
message   in   email   is   action   oriented   (e.g.   when   people   send   us   FYIs)   and   not   every   web   page   is   purely  
informative  (e.g.  when  it  contains  a  request  to  complete  a  form).    

This  distinction  has  critical  implications  for  how  we  treat  personal  information.  For  reasons  that  
will   become   clear,   it   is   often   impossible   to   discharge   action-­‐oriented   items   immediately.   So   reminding  
strategies  (e.g.  creating  task  related  email  folders,  or  leaving  active  documents  on  the  desktop),  have  to  
be  set  up  to  prompt  the  user  about  their  commitments  with  respect  to  the  undischarged  item.  Failure  to  
set  up  such  structures  can  have  severe  implications  for  job  success  and  productivity;  we  mustn’t  forget  to  
respond  to  that  important  request  from  our  boss,  even  when  we  are  inundated  with  other  commitments.  
In  contrast,  how  we  deal  with  informative  items  is  usually  more  discretionary:  they  usually  do  not  need  to  
be  actively  processed  to  meet  deadlines,  so  it  is  less  critical  that  we  create  dedicated  reminding  structures  
to  ensure  that  they  are  dealt  with  appropriately.    

Information  Uniqueness  

Another  critical  information  property  is  uniqueness.  Uniqueness  has  strong  implications  for  how  
we   deal   with   personal   information.   Certain   types   of   information   (such   as   personal   files   that   we   create  
ourselves)  may  be  resident  only  on  our  computer.  As  a  result  we  may  be  the  only  person  in  the  world  who  
 8  

has  access  to  those  items.  Those  who  have  lost  data  following  system  crashes  are  only  too  aware  that  if  
we   do   not   take   responsibility   for   storing   and   maintaining   unique   data,   then   it   will   not   be   preserved   for  
future   access   (Marshall,   2008a,b).   In   contrast,   public   information   such   as   web   data   may   be   resident   on  
multiple  servers  and  may  be  recoverable  even  if  we  personally  take  no  action  to  store  a  local  copy.  Email  
data   lies   somewhere   in   between.   We   may   be   able   to   ask   coworkers   to   regenerate   a   copy   of   an   important  
message  that  we  have  temporarily  mislaid,  or  lost  in  a  system  crash,  but  we  can’t  guarantee  they  will  have  
kept  that  information.    

It   is   important   to   note   that   uniqueness   is   defined   subjectively:   relative   to   our   own   goals   and  
interests.   There   are   innumerable   unique   information   items   in   the   world,   but   as   curators   we   are   only  
concerned   to   take   decisive   action   to   preserve   those   that   are   relevant   to   us.   Other   people’s   information  
may   be   equally   important   to   them,   but   there   is   no   reason   why   we   should   be   concerned   to   preserve   it,  
unless  of  course  we  work  with  them.  This  personal  uniqueness  is  often  associated  with  information  that  
we   have   invested   effort   in   generating.   If   we   have   dedicated   substantial   time   in   generating   an   information  
item   (e.g.   an   extended   personal   document,   a   carefully   crafted   presentation,   or   a   collection   of   wedding  
photos),   then   that   information   will   be   something   that   we   make   enormous   efforts   to   preserve,   in   part  
because  of  the  effort  involved  in  regenerating  it.      

Uniqueness  has  a  huge  impact  on  our  management  strategies.  No-­‐one  else  will  take  care  of  our  
unique   personal   data.   We   personally   need   to   create   reliable   structures   for   re-­‐accessing   highly   personal  
data   such   as   passwords,   tax   forms,   passport   details   or   financial   records,   even   when   we   rarely   need   to  
access  this  information.  Personally  produced  documents  also  tend  to  be  unique  and  need  to  be  carefully  
organised.   The   same   is   largely   true   for   emails:   we   need   to   have   reliable   methods   for   reaccessing   these  
because  we  cannot  always  rely  on  others  to  keep  the  important  messages  that  we  need.  Web  pages  are  
rather  different.  They  are  generally  more  easily  recoverable  (via  search  or  browsing)  even  if  we  have  not  
bookmarked   them.   And   in   addition,   because   we   have   not   usually   been   responsible   for   generating   their  
content,  we  are  not  as  concerned  if  we  cannot  recover  the  information  they  contain.    

Information Type Action vs. Information Uniqueness

oriented?
Personal paper documents Action oriented if self created and Unique if self created or
current annotated
Long term archives tend to be
informational
Personal electronic documents Action oriented if self created and Unique if self created
current
Long term archives tend to be
informational
Email Often action oriented Range from unique to non-unique
Long term archives tend to be mass mailings
informational
Personal photos Affective Predominantly unique
Web Informational Non-unique

Table  2:  Main  properties  of  different  information  types.  

 9  

Table   2   shows   the   key   properties   of   common   classes   of   information,   such   as   paper,   electronic  
files,   email,   photos   and   web   documents.   The   Table   shows   these   are   very   different   with   respect   to   action-­‐
orientation   and   also   for   uniqueness.   Current   paper   and   electronic   documents   and   emails   are   often   action  
oriented.  Paper  and  electronic  documents  and  personal  photos  tend  to  be  unique.  These  differences  have  
strong  implications  for  curation.  The  uniqueness  of  paper  or  electronic  personal  documents,  and  personal  
photos   leads   people   to   be   very   conservative   and   to   keep   most   of   these   items.   They   also   have   to   preserve  
action   oriented   items   such   as   emails,   and   personal   documents   in   such   a   way   that   promotes   effective  
action.        

We  now  turn  to  each  of  the  main  curation  processes,  describing  how  people  Keep,  Manage  and  
Exploit   their   personal   information.   It   will   be   clear   from   our   prior   discussion   that   there   are   huge  
dependencies  between  these  processes.  In  what  follows,  we  review  each  of  these  processes  separately,  
but  we  should  not  lose  sight  of  the  relationships  between  them.    

3. KEEPING    

3.1  OVERVIEW ,  PROBLEMS  AND  STRATEGIES  

We  encounter  too  much  information  to  keep  it  all,  because  of  various  costs  including:  

- Management  costs  –  we  need  to  organise  information  if  we  are  to  obtain  value  from  it.  The  more  
we   keep,   the   more   management   effort   is   required.   Some   visions   of   new   technology   suggest   that  
in  the  future  our  information  will  be  organised  automatically,  but  these  technologies  are  not  yet  
in  place.  Indeed,  in  future  sections,  we  explore  whether  these  technologies  will  ever  effectively  
replace  the  need  for  manual  organisation.  
 
- Exploitation  costs  –  keeping  information  of  low  value  increases  the  difficulty  of  retrieval.  Keeping  
too   many   items   can   be   distracting   if   manual   browsing   is   used   for   access.   Nevertheless,   there   are  
those  who  contend  that  future  retrieval  will  be  entirely  search  based  reducing  exploitation  costs  
regardless  of  how  much  we  keep.  
 

Keeping  decisions  are  a  fact  of  life.  Every  day  we  receive  new  emails,  create  new  files  and  folders  
and   browse   new   web   sites.   Some   of   this   information   is   of   little   long-­‐term   value,   but   some   of   it   is   task  
critical   and   needs   to   be   preserved   for   the   long   term.   Data   extracted   from   Boardman   and   Sasse   (2004)  
1
suggests  that  users  acquire  an  average  of  5  new  files  per  day,  5  emails   per  day,  and  one  bookmark  every  
5   days.   Other   studies   indicate   people   acquire   one   new   contact   per   day   (Whittaker   et   al.,   2002a),   and  
around   5   digital   photos   (Whittaker   et   al.,   2010).   But   these   statistics   are   an   over-­‐simplification   of   the  
complexity   of   keeping   decisions.   The   statistics   record   positive   decisions,   but   fail   to   register   the   many  
decisions   to   reject   information   judged   to   be   of   little   value.   To   be   more   specific,   we   know   that   users  
receive   an   average   of   42   emails   per   day;   so   focusing   exclusively   on   what   they   actively   decide   to   keep  
overlooks   the   37   decisions   they   make   to   delete   irrelevant   information.   For   email   alone,   making   the   highly  
                                                                                                                                   
1
Although  Dabbish  et  al.  (2005)  suggest  higher  keeping  rates  for  email.  
 10  

conservative  assumption  that  email  volumes  will  not  change  over  our  lifetimes,  this  equates  to  around  1  
million  keeping  decisions  over  a  60  year  digital  life.  

We  know  from  various  interview  and  survey  studies  how  difficult  people  find  it  to  decide  what  
information   they   want   to   keep   (Bergman   et   al.,   2009,   Boardman   and   Sasse,   2004,   Jones,   2004,   2007,  
Whittaker  and  Hirschberg,  2001,  Whittaker  and  Sidner,  1996).  But  why  are  keeping  decisions  so  difficult?  
One  reason  is  that  they  require  us  to  predict  the  future.  To  decide  what  to  keep,  we  have  to  determine  
the  probable  future  value  of  an  information  item.    

This   may   be   a   general   psychological   problem.  There   is   a   great   deal   of   psychological   research   that  
shows  that  people  are  poor  at  making  many  types  of  decisions  that  involve  their  future.  Such  prediction  
requires   people   to   reason   about   hypothetical   situations,   which   they   are   notoriously   poor   at.   People’s  
predictions   are   also   subject   to   various   types   of   bias.   For   example,   they   expect   the   future   to   be   very   much  
like  the  present,  and  their  predictions  are  unduly  influenced  by  recent,  or  easily  recalled,  events  (Gilbert,  
2006,  Kahneman  and  Tversky,  1979).    

In  the  PIM  context,  the  keeping  decision  requires  people  to  predict  future  informational  contexts  
and   assess   future   informational   requirements.   In   deciding   what   to   keep,   people   have   to   evaluate   the  
potential   future   utility   of   keeping   an   item,   and   weigh   this  against   potential   management   and   exploitation  
costs  associated  with  keeping  it.  Jones  (2004)  argues  that  the  decision  whether  “to  keep  or  not  to  keep”  
information  for  future  usage  is  prone  to  two  types  of  costly  mistakes.  On  the  one  hand  information  not  
kept  is  unavailable  when  it  is  needed  later.  On  the  other,  keeping  irrelevant  information  not  only  causes  
guilt  about  being  disorganized  (Boardman  and  Sasse,  2004,  Whittaker  and  Sidner,  1996),  it  also  increases  
retrieval  time.  Irrelevant  information  competes  for  the  user’s  attention,  obscuring  important  information  
relevant   to   the   current   task.   Indeed   it   is   well   known   in   psychology   that   in   visual   search   the   number   of  
irrelevant  distracters  increases  the  time  taken  for  people  to  identify  a  target  object  (Treisman  and  Gelade,  
1980).  Furthermore,  there  is  a  “deletion  paradox”:  while  unimportant  information  items  distract  attention  
and  increase  retrieval  time  for  important  items,  it  takes  time  and  effort  to  review  items  to  decide  whether  
to  delete  them  (Bergman  et  al.,  2009).    

When   people   weigh   up   the   advantages   of   keeping   versus   deleting,   some   of   the   reasons   for  
keeping   are   rational   -­‐   after   all   the   user   can   always   think   of   a   situation   when   the   information   item   may  be  
needed   (Whittaker   and   Hirschberg,   2001).   However   there   are   also   less   rational   reasons   why   people   avoid  
deletion,   which   can   be   attributed   to   general   psychological   decision   making   processes   (Kahneman   and  
Tversky,   1979).   In   making   decisions,   losses   and   gains   are   evaluated   asymmetrically:   losses   are   more  
salient  than  gains,  and  the  possible  loss  of  information  emotionally  influences  the  decision  maker  more  
than   the   gains   of   reduced   retrieval   time.   And   small  objective  probabilities  are  subjectively   weighted   more  
highly   than   their   actual   likelihood.   Thus   people   perceive   as   significant   the   very   small   probability   that   a  
deleted  information  item  will  be  needed.      

We   now   review   various   studies   looking   at   people’s   keeping   decisions,   when   managing   their  
paper  archives,  emails,  contacts,  web  pages  and  personal  photos.    

3.2  KEEPING  PAPER    

Somewhat   curiously,   despite   the   prevalence   of   keeping   decisions,   there   have   been   relatively   few  
studies  that  have  looked  directly  at  this.  One  exception  is  a  study  of  people’s  paper  archiving  behaviour  
 11  

(Whittaker   and   Hirschberg,   2001).   While   there   is   a   common   intuition   that   the   world   is   shifting   away   from  
paper  and  becoming  more  digital,  we  will  see  that  people  treat  paper  in  ways  that  are  very  similar  to  their  
treatment  of  digital  information.    

One   methodological   problem   with   investigating   keeping   behaviour   lies   in   finding   contexts   where  
people  are  explicitly  focused  on  the  keeping  decision.  Our  study  identified  one  such  situation.  Participants  
were   about   to   move   offices   and   had   to   make   decisions   about   which   information   to   keep   and   what   to  
throw   away.   When   we   interviewed   them,   they   had   all   recently   sorted   through   their   paper   archives   in  
preparation  for  the  move.  Their  new  offices  had  reduced  personal  storage  space  compared  their  existing  
offices,  although  extra  storage  was  provided  in  public  locations.  This  reduction  in  local  storage  motivated  
careful   reflection   as   well   as   sorting   and   discarding   existing   data.   In   interviewing   and   surveying   workers  
when   we   did,   we   capitalised   on   the   fact   that   they   had   very   recently   handled   most   of   their   paper   data,  
forcing  them  to  identify  criteria  for  determining  what  to  keep  and  what  to  discard.  

Discarding  Behaviour  

People   experienced   major   problems   in   deciding   what   to   keep   and   what   to   throw   away.   As   the  
psychology   literature   would   suggest,   there   was   a   bias   towards   preservation.   Even   after   spending   large  
amounts   of   time   deciding   what   to   discard,   workers   still   retained   huge   archives   after   the   move.   In  
preparation  for  the  move,  people  spent  almost  nine  hours  rationalising  their  data,  and  reported  that  this  
was  a  difficult  process.  Despite  these  efforts,  the  final  amount  of  information  that  people  actually  threw  
away  was  small  compared  with  what  they  kept:  people  discarded  just  22%  of  their  original  archives,  with  
the  final  preserved  archive  on  average  being  more  than  18  mover’s  boxes  (equivalent  in  volume  to  a  pile  
of  telephone  directories  about  30m  high).    

We  looked  at  the  characteristics  not  only  of  what  people  kept,  but  also  what  was  discarded.  As  
we  expected,  at  least  part  of  discarded  data  was  once  valuable  information  that  had  become  obsolete.  As  
jobs,   personal   interests   or   company   strategy   changes,   then   the   value   of   particular   information   decreases.  
But  not  all  discarded  information  underwent  the  transition  from  valuable  to  obsolete.  For  example,  23%  
of  discarded  data  was  unread.  Why  would  people  keep  information  that  they  had  never  even  looked  at?  
Two   general   problems   led   to   this   accumulation   of   superfluous   information.   First,   people   experience  
problems  with  information  overload  leading  them  to  only  partially  process  incoming  information.  Second  
they   engage   in   deferred   evaluation   of   what   to   keep   -­‐   causing   them   acquire   large   amounts   of   data   that  
later  turn  out  to  be  extraneous.  

Information   overload   refers   to   the   fact   that   people   have   insufficient   time   to   process   all   the  
information   they   are   exposed   to.   One   consequence   of   information   overload   is   that   non-­‐urgent  
information   is   never   processed.   Non-­‐urgent   data   are   set   aside   (often   in   optimistically   named   “to   read”  
piles),   accumulating   indefinitely,   because   the   same   time   pressures   that   prevent   complete   processing   of  
incoming  data  also  prevent  rationalising  (“clean-­‐up”)  of  archives.  Consequently,  people  seldom  discover  
that   their  unread  non-­‐urgent  documents  are  superfluous  until  exceptional  circumstances  (like  the  current  
office  move)  force  people  to  scrutinise  their  archives.  

Yet  even  when  people  find  the  time  to  systematically  examine  new  information,  uncertainty  of  
the  future  value  of  that  information  means  they  are  often  highly  conservative:  postponing  final  judgments  
about   utility   until   some   unspecified   future   date.   Some   people   deliberately   defer   evaluation   about  
 12  

incoming  information,   allowing  time  to  pass  so  as  to  make  better-­‐informed  judgments  about  information  
utility.  Often  these  post-­‐hoc  judgements  are  based  on  whether  information  was  ever  actually  used.    

Deferred  evaluation  means  people  retain  information  of   unclear  value  -­‐  just  in  case  it  later  turns  
out  to  be  useful.  Finally,  judgments  about  potential  utility  are  made  more  difficult  because  the  value  of  
data  can  change  over  time.  Knowing  that  the  value  of  information  might  change  also  leads  some  people  
to  postpone  the  keeping  decision  while  there  is  still  archival  space.  

Accumulating   unprocessed   data   and   deferring   evaluation   are   good   from   the   (conservative)  
perspective  that  potentially  valuable  information  is  not  lost.  However  the  problem  with  this  approach  is  
that   people   seldom   revisit   their   archives   to   rationalise   them,   so   their   archives   end   up   containing  
considerable  amounts  of  information  of  dubious  value.  Thus,  74%  of  our  users  had  not  cleaned  out  their  
archives  for  over  a  year.  Furthermore,  very  few  clean-­‐ups  occur  spontaneously:  84%  arise  from  extrinsic  
events   such   as   job   changes   or   office   moves.   This   infrequency   of   clean-­‐ups   means   that   documents   are  
often  not  discovered  to  be  superfluous,  until  they  have  been  stored  for  some  time.    

To   sum   up,   our   deletion   data   illustrates   important   aspects   of   keeping.   When   extraordinary  
events  such  an  office  move  occur,  then  people  discard  about  22%  of  their  data,  some  of  which  is  obsolete.  
However   other  factors   besides   obsolescence   such   as   information   overload   and   deferred   evaluation   mean  
that  archives  are  polluted  by  marginally  relevant  data.  Rather  than  discarding  once-­‐valuable  information  
that   is   now   of   little   utility,   much   of   what   people  later   discard   is   unprocessed   information   they   have   never  
properly  evaluated,  or  kept  ‘just  in  case’.  

What  do  we  keep  and  why  do  we  keep  it?    

We  also  looked  at  the  properties  of  the  information  people  kept,  and  their  reasons  for  keeping  it.  
One  conjecture  was  that  a  large  proportion  of  the  information  that  people  kept  would  be  unique  to  that  
person;   because   other   people   will   not   take   responsibility   for   retaining   highly   personal   data.   In   contrast,  
we  expected  people  to  be  much  less  likely  to  keep  publicly  available  data.  Why  take  responsibility  for  data  
that  is  available  elsewhere?  

Uniqueness   was   clearly   important   in   determining   whether   users   would   preserve   certain  
documents.   Unique   data   are   usually   highly   associated   with   their   archiver.   Three   types   of   unique   data  
accounted   for   49%   of   people’s   archives:   working   notes,   archives   of   completed   projects,   and   legal  
documents.    

But  contrary  to  our  expectations,  uniqueness  was  not  the  sole  criterion  for  deciding  to  keep  data.  
Only  49%  of  people’s  original  archive  was  unique:  15%  was  unread,  but  36%  consisted  of  copies  of  publicly  
available   documents.   We   have   already   discussed   why   people   preserve   unread   data,   but   why   keep  
duplicates   of   public   documents?   Four   main   reasons   were   given:   availability,   reminding,   lack   of   trust   in  
external  stores,  and  sentiment.    

Availability   allows   relevant   materials   to   be   at   hand   when   people   need   them.   Several   people  
mentioned  not  wanting  to  experience  the  delay  associated  with  refinding  information,  or  even  accessing  
it   on   the   Web.   In   other   words   they   wanted   to   reduce   their   exploitation   costs   by   keeping   valued  
information  in  a  personal  archive.  
 13  

Reminding   relates   to   availability.   A   personal   copy   prompts   people   about   outstanding   actions  
associated   with   a   document,   or   simply   reminds   them   they   are   in   possession   of   that   information.  
Documents   in   public   or   digital   stores   seemed   less   capable   of   supporting   reminding.   People   also   kept  
personal   copies   of   public   data   because   they   didn’t   trust   other   archival   institutions   to   keep   the   documents  
they  needed.  Distrust  of  external  stores  also  extended  to  digital  resources  such  as  the  Web.    

In   addition   to   these   functional   reasons,   people   described   sentimental   reasons   for   keeping  
information.   People   admit   such   information   has   little   relevance   for   likely   future   activities,   but   they   still  
cannot  part  with  it,  because  it  is  part  of  their  intellectual  history  and  professional  identity.    

Another  potential  reason  for  keeping  personal  copies  of  publicly  available  documents  is  that  they  
contain   personal   annotations.   Other   research   has   documented   the   utility   of   annotations   for   focusing  
attention  and  improving  comprehension  of  what  is  read  or  heard  (Kalnikaite  and  Whittaker,  2007,  2008,  
Sellen  and  Harper,  2002).  Although  most  people  made  such  annotations,  they  seemed  of  little  long-­‐term  
use,  however.  Many  people  stated  that  annotations  had  transient  value,  becoming  uninterpretable  after  
some  time  has  elapsed.  This  is  consistent  with  recent  studies  of  long  term  note-­‐taking  showing  that  the  
utility  of  handwritten  notes  decreases  even  after  a  month  (Kalnikaite  and  Whittaker,  2007,  2008).    

3.3  KEEPING  EMAIL    

Email   is   different   from  either  self  created   files,   or   documents   accessed  on  the  web.  One  major  
difference  is  that  a  significant  proportion  of  the  information  we  receive  in  email  is  actionable,  i.e.  we  have  
to   respond   to   it   or   process   it,   often   within   a   specific   time   frame.   This   contrasts   with   most   web-­‐based  
information   which   does   not   demand   an   action.   Another   significant   contrast   to   self-­‐created   files   is   that  
most   emails   are   generated   by   others   (who   in   some   cases   are   unfamiliar   to   the   main   user).   This   lack   of  
familiarity  sometimes  makes  it  harder  for  people  to  decide  on  the  utility  of  such  email  information.  A  final  
rather  different  characteristic  of  email  is  its  sheer  variability.  In  our  inboxes  we  may  see  many  different  
types   of   messages   including:   tasks   or   todo   items,   documents/attachments,   fyis,   schedules,   social  
messages   and   jokes.   Again   this   heterogeneity   makes   the   keeping   decision   rather   different   from   other  
information  types.    

Overall   we   keep   about   70%   of   our   emails   (Dabbish   et   al.,   2005).   This   seems   a   surprisingly   high  
retention   rate   given   the   apparent   irrelevance   of   many   of   the   emails   we   receive,   but   there   are   reasons   for  
this.   In   what   follows,   we   separately   discuss   people’s   keeping   behaviours   for   informational   versus  
actionable  messages,  as  keeping  behaviour  is  very  different  for  each  of  these.    

Informational  messages  

Informational   messages   form   about   one   third   (34%)   of   what   is   delivered   in   email   (Dabbish   et   al.,  
2005).   Informational   messages   are   treated   in   a   similar   manner   to   paper   documents.   As   with   paper,   the  
keeping   decision   is   often   difficult.   People   find   it   hard   to   judge   the   value   of   incoming   informational  
messages,   so   they   use   the   deferral   strategy.   Rather   than   investing   valuable   time   in   reading   a   new  
informational  message,  users  register  its  arrival,  but  defer  dealing  with  it  until  they  are  more  certain  of  its  
value.  Deferred  emails  are  ‘kept  around’  allowing  more  informed  judgements  to  be  made  later.    
 
Users  are  aware  that  deferred  messages  need  to  be  re-­‐evaluated  at  a  later  point.  Some  employ  
folders   for   this   purpose:   and   28%   of   informational   messages   are   filed   for   later   reading   (Dabbish   et   al.,  
 14  

2005).  However  the  problem  with  this  strategy  is  that  filing  may  lead  these  messages  to  be  ‘out  of  sight  
and   out   of   mind’   as   such   folders   are   seldom   revisited   (Whittaker   and   Sidner,   1996).   A   more   common  
strategy  is  to  leave  them  in  the  inbox:  Dabbish  et  al.  found  42%  of  informational  messages  are  left  in  the  
inbox,  to  increase  the  probability  that  deferred  evaluation  will  actually  take  place.  The  inbox  is  an  active  
workspace:   leaving   information   there   increases   the   chance   that   information   will   be   re-­‐visited   as   users  
reaccess   the   inbox   to   process   incoming   messages.   But   there   is   an   obvious   downside   to   this   strategy.  
Although   the   strategy   increases   the   probability   of   revisiting   ‘yet   to   be   decided   items’,   the   presence   of  
such   unevaluated  information  makes  it  more  difficult   for   people   to  locate   important   information,  such  as  
messages  requiring  action  (Bellotti  et  al.,  2003,  Whittaker,  2005,  Whittaker  and  Sidner,  1996).    

As   with   paper   archives,   people   experience   information   overload   in   email.   Overload   may   lead  
people   to   defer   completely   reading   each   message   until   they   have   more   time.   And   of   course   because   they  
are  constantly  bombarded  with  more  incoming  messages,  they  often  never  return  to  deferred  messages  
(Whittaker   and   Sidner,   1996).   One   factor   contributing   to   whether   a   message   is   read   or   not   is   its   length  
and   Whittaker   and   Sidner   (1996)   found   that   21%   of   inbox   messages   contained   more   than   5   screenfulls   of  
text,  consistent  with  the  fact  that  people  leave  longer  messages  there  for  later  reading.    
 
Actionable  Items  

Actionable   messages   are   those   that   we   have   to   do   something   specific   about.   In   an   ideal   world  
(such   as   that   inhabited   by   management   consultants),   we   might   process   these   messages   just   once,  
carrying  out  the  required  action  and  then  deleting  them.  This  is  often  referred  to  as  the  one  touch  model.  
The   advantages   of   the   model   are   obvious:   touching   a   message   just   once   means   that   users   don’t   forget   to  
deal   with   it,   and   they   don’t   have   to   repeatedly   reconstruct   the   context   of   old   messages   when   they  
eventually  come  to  process  these.  And  if  messages  are  processed  at  once  this  keeps  the  inbox  clear  for  
important  incoming  messages.      

Some   users   try   to   adhere   to   this   model:   overall   users   reply   to   65%   of   actionable   messages  
immediately  (Dabbish  et  al.,  2005).  An  immediate  reply  clearly  reduces  the  chance  that  they  will  forget  to  
act   on   a   message.   However   even   when   people   do   reply   immediately   they   still   keep   85%   of   actionable  
messages,  suggesting  that  one  touch  does  not  describe  actual  practice.    

There  are  several  reasons  for  such  retention.  In  some  cases,  one  touch  and  an  immediate  reply  
are   not   possible.   Many   important   email   tasks   are   too   complex   or   lengthy   be   executed   immediately  
(Bellotti   et   al.,   2005,   Venolia   at   al.,   2001,   Whittaker   and   Sidner,   1996,   Whittaker,   2005).   This   leads   to  
deferral  of  37%  of  actionable  messages  (Dabbish  et  al.,  2005).  Deferral  is  often  a  direct  consequence  of  
interdependent  tasks,  i.e.  those  collaborative  tasks  involving  tight  collaboration  with  others  (Bellotti  et  al.,  
2005,  Whittaker,  2005).  Interdependence  results  in  both  iteration  and  delays  between  messages  relating  
to   the   task.   Iteration   arises   because   interdependent   tasks   often   require   multiple   exchanges   between  
participants   (Bellotti   et   al.,   2005,   Venolia   et   al.,   2001,   2003,   Whittaker   and   Sidner,   1996).   People   may  
need   to   negotiate   exactly   what   a   collaborative   email   task   involves,   or   who   will   be   responsible   for   each  
component.   This   consensus   needs   to   be   built   and   multiple   responses   often   need   to   be   collated.   Delays  
occur   because   these   negotiations   take   time   and   because   collaborators   often   lack   the   necessary  
information  to  respond  immediately  to  address  their  part  of  the  task.  One  way  to  estimate  the  prevalence  
of   interdependent   tasks   is   by   determining   how   many   emails   are   part   of   a   conversational   thread,   as  
 15  

threads   indicate   relations   and   common   underlying   activities   among   messages.   Threading   estimates   range  
from  30-­‐62%  of  messages  (Bellotti  et  al.,  2003,  Whittaker  et  al.,  2007).  

The   need   for   deferral   of   actionable   messages   has   important   consequences   for   keeping.   Unless  
actions   are   discharged,   messages   are   usually   ‘kept   around’   as   reminders   that   they   are   still   incomplete.  
Actionable   messages   are   therefore   almost   always   kept   (only   0.5%   are   deleted).   This   figure   is   much   higher  
than  for  information  messages,  30%  of  which  are  deleted.  Furthermore,  actionable  messages  have  to  be  
kept  in   a  way   that  guarantees  that  they  will  be  reencountered.  It’s  no  good  deferring  ‘todo’  emails,  unless  
you   have   some   method   of   guaranteeing   that   you   actually   return   to   them.   We   revisit   this   issue   in   the   next  
section,  when  we  talk  about  management  strategies.    

3.4  KEEPING  CONTACTS  

Contact  management  is  another  area  that  demands  careful  keeping  decisions.  Whittaker  et  al.,  
(2002a)  looked  at  people’s  address  books,  rolodexes,  calendars  and  contact  management  programs  and  
explored  the  criteria  that  people  used  for  including  someone  in  their  contact  list.  We  are  overloaded  with  
respect  to  the  contacts  we  encounter.  We  are  cced  on  many  messages,  and  we  read  web  pages  or  blogs  
from   friends,   colleagues   and   strangers.   Some   of   these   are   people   who   we   want   to   interact   with   again.  
Others  may  have  been  involved  in  one-­‐off  conversations  that  require  no  follow-­‐up.  Contact  management  
requires  decisions  about  which  people  you  decide  to  keep  contact  information  about,  as  well  as  the  types  
of  information  that  you  decide  to  keep  about  those  people.    

It  is  complex  to  decide  on  important  contacts  from  the  many  people  that  you  are  exposed  to  on  a  
daily   basis.   As   with   paper   and   email   archives,   it   is   hard   to   anticipate   whether   you   will   need   to  
communicate  with  that  person  in  the  future:  whether  someone  is  an  “important  contact”  becomes  clear  
only  over  time.   Just  as  with  the  deferral  strategy,  our  informants  often  “over-­‐saved”  information,  leading  
to   huge   rolodexes,   overflowing   booklets   of   business   cards,   and   faded   post-­‐it   notes   scattered   around   their  
work   areas.   But   despite   this   strategy,   participants   were   exposed   to   many   more   contacts   than   they  
recorded  information  about.    

We   identified   specific   factors   that   were   critical   in   determining   important   contacts.   Just   as   with  
deferred   evaluation   in   email   and   paper   archives,   the   final   decision   to   keep   depends   on   past   interaction  
with  the  contact,  in  particular   frequency  and  recency  of  communication.  People  also  noted  how  difficult  it  
was  to  make  decisions  about  the  future,  based  on  short  term  interactions  and  scanty  evidence.   Again  we  
see   the   importance   of   long   term   information   in   evaluating   contacts:   important   contacts   are   those   with  
whom   we   have   repeated   interactions   over   extended   periods.   In   addition   the   selection   process   is   error-­‐
prone,   because   of   the   difficulty   of   predicting   long-­‐term   relationships   on   the   basis   of   brief   initial  
interactions.    

In   a   follow   up   study,   we   presented   people   with   contacts   mined   from   their   email   archives,   and  
asked   them   to   distinguish   between   important   and   unimportant   ones.   The   findings   were   quite   striking.  
Despite  having  huge  archives  of  contacts  (858  on  average),  participants  rated  only  14%  (118)  as  important  
and   ‘worth   keeping’.   Criteria   for   inclusion   echoed   those   identified   in   our   earlier   interviews:   participants  
chose   contacts   with   whom   they   interacted   frequently   and   recently,   as   well   as   for   a   long   time,   and   who  
were  likely  to  respond  to  their  emails.  They  also  excluded  spammers.    
 16  

Overall  there  are  interesting  parallels  between  contacts,  paper  and  emails.  People  are  exposed  
to  many  more  contacts  than  they  can  record  systematic  information  for,  so  they  defer  making  decisions  
managing  to  reserve  judgment  and  ‘overkeep’  data  about  contacts  that  they  don’t  need.  Furthermore,  the  
criteria   that   people   use   to   judge   the   value   of   contacts   are   based   around   usage   and   interaction:   valued  
contacts   are   those   who   are   interacted   with   often,   frequently   and   recently.   However   one   key   difference  
between   contacts,   email   and   papers   is   that   users   ignore   or   ‘discard’   a   much   higher   percentage   of  
encountered  contacts.    

3.5  KEEPING  WEB  PAGES  

Similar  problematic  keeping  decisions  also  surface  on  the  web  (Jones,  2004),  where  we  see  errors  
of  commission  (over-­‐keeping  information  that  turns  out  to  have  little  future  value)  and  omission  (failing  to  
keep   information   that   turns   out   to   be   needed   later).   There   are   clear   errors   of   commission;   e.g.   people  
expend   energy   creating   bookmarks   that   they   never   subsequently   use.   Tauscher   and   Greenberg   (1997)  
showed  that  58%  of  bookmarks  people  are  never  used,  suggesting  poor  decision  making.    

At   the   same   time,   other   studies   of   web   behaviours   reveal   failures   of   omission   -­‐   where   people  
don’t   preserve   information   that   does   turn   out   to   be   useful   later.   Wen   (1993)   coined   the   term   post  
retrieval   value   to   describe   web   resources   that   people   have   accessed   but   not   preserved   -­‐   only   later  
realising   their   utility.   His   study   showed   that   people   were   only   able   to   later   find   about   20%   of   information  
they   have   previously   accessed   and   attended   to,   in   an   earlier   information   retrieval   session.   Such   failure  
partially   originates   from   an   unwillingness   to   make   deliberate   attempts   to   keep   information;   his   users  
were   unwilling   to   create   bookmarks   as   records   of   useful   pages,   because   these   would   ‘clutter’   up   their  
current   bookmark   collection.   These   findings   were   replicated   in   other   similar   studies   (Aula   et   al.,   2005).  
Instead  users  preferred  to  try  to  retrace  their  original  searches  –  a  strategy  which  is  often  unsuccessful.  

3.6  KEEPING  PHOTOS  

With  the  advent  of  digital  photography,  the  numbers  of  pictures  that  people  are  now  taking  has  
increased  massively  (Bentley  et  al.,  2006,  Kirk  et  al.,  2007,  Whittaker  et  al.,  2010,  Wilhelm  et  al.,  2004),  
and  similar  keeping  issues  are  beginning  to  arise  for  digital  photos.  Now  that  people  have  collections  of  
thousands  of  digital  pictures,  how  do  they  decide  which  of  these  to  keep  and  which  to  delete?  

We  looked  at  this  in  a  study  of  parents  with  young  families  (Whittaker  et  al.,  2010)  who  had  an  
average  of  4475  digital  pictures.  All  participants  deleted  some  pictures,  both  when  pictures  were  taken,  
and  when  they  were  uploading  from  picture  to  camera.  Participants  estimated  they  deleted  on  average  
17%   of   their   pictures.   The   reasons   people   gave   for   deletion   were   that   the   pictures   were   poor   technical  
quality  or  did  not  capture  the  event  of  interest.  In  general,  deletion  was  a  difficult  process,  as  evidenced  
by   the   fact   that   many   of   the   pictures   that   were   kept   were   near   duplicates   (i.e.   multiple   pictures   of  
identical   scenes),   an   observation   that   is   confirmed   in   other   studies   (Kirk   et   al.,   2007),   suggesting   that  
people  are  keeping  their  options  open  about  the  best  view  of  a  given  scene.  One  of  the  reasons  people  
gave   for   this   ‘overkeeping’   was   that   they   perceived   little   cost   in   keeping   many   photos.   They   weren’t  
therefore  focused  on  the  exploitation/retrieval  context  when  they  made  keeping  decisions.  As  with  paper  
and   email,   when   we   probed   people   further   about   this   conservative   approach   to   keeping,   people   had   a  
strong   expectation   that   they   would   return   to   their   photo   collection   to   rationalise   it   at   a   later   date.   And   as  
in  our  paper  and  email  studies,  this  rationalisation  seldom  occurred.  
 17  

3.7  KEEPING  SUMMARY  

1. Keeping   decisions   are   difficult   because   they   require   people   to:   (a)   predict   their   future   retrieval  
needs,   (b)   take   into   account   the   possibility   that   those   information   needs   may   change,   and   (c)  
make  utility  decisions  under  conditions  of  information  overload,   often  on  incomplete  readings  of  
information.  
2. Errors  are  made:  the  primary  tendency  is  overkeeping,  i.e.  keeping  things  that  are  never  accessed  
(observed  with  paper,  email,  contacts  and  photo  archives),  although  there  is  evidence  from  some  
web  studies  of  failing  to  keep  information  that  later  turns  out  to  be  relevant.  
3. Consistent   with  overkeeping,   deletion   is   relatively   infrequent,   varying   between   around   17%   for  
photos  to  30%  for  emails.  Contacts  are  very  different,  however,  it  seems  that  because  people  are  
exposed  to  many  more  of  these,  they  are  happy  to  ignore  86%  of  the  contacts  they  encounter.  
4. The   nature   of   the   information   item   affects   the   keeping   decision.   This   decision   is   relatively  
straightforward  for  certain  items:  we  obviously  need  to  keep  unaddressed  actionable  emails  or  
unique   personally   generated   items   that   no-­‐one   else   will   safeguard.   However   it’s   very   hard   for  
people  to  decide  the  value  of  data  such  as  public  web  pages  or  informational  emails.  
5. Because  of  problems  in  making  the  keeping  decision,  rather  than  viewing  keeping  as  a  one-­‐time  
decision,   people   often   used   a   deferral   strategy   –   waiting   to   see   whether   information   is   useful.  
Two  major  weaknesses  of  deferral  are  (a)  that  people  seldom  return  to  their  collections  to  carry  
out   a   re-­‐evaluation   of   tentatively   kept   information;   (b)   deferral   means   that   collections   are   full   of  
items  of  dubious  value  -­‐  that  make  it  more  difficult  to  find  truly  valuable  information.      
6. People  don’t  generally  seem  to  be  aware  of  the  implications  of  overkeeping.  While  they  complain  
about   how   full   their   inboxes   are,   they   nevertheless   delete   only   30%   of   emails,   and   even   after  
spending   days   working   though   paper   archives   they   still   preserve   78%   of   those.   On   the   web,   in  
contrast,  there  is  a  suggestion  that  people  don’t  bookmark  because  they  are  aware  that  this  will  
make  valued  materials  harder  to  find.  This  could  be  because  they  consider  web  information  to  be  
unimportant  or  because  they  think  it  is  easily  recoverable  by  other  means.    

4. MANAGEMENT  

4.1  OVERVIEW ,  PROBLEMS  AND  STRATEGIES  

We   will   first   describe   different   methods   for   organising   information,   as   well   as   the   trade-­‐offs  
between   these.   We   next   discuss   factors   which   influence   users’   choice   of   management   strategies   and  
studies  evaluating  the  utility  of  these  different  strategies.  We  then  briefly  talk  about  a  radical  alternative  
which   proposes   that   we   forgo   preparatory   organisation   altogether   and   rely   totally   on   search   for  
information  exploitation.    

Management   is   a   crucial   curation   process   because   it   directly   affects   exploitation.   We   are  

constantly   acquiring   information,   and   over   long   periods   large   amounts   of   personal   information   clearly  
accumulate   (Marshall,   2008a,b).   Using   current   estimates   of   how   many   documents,   digital   photos   and  
emails   we   acquire   on   a   daily   basis   (Boardman   and   Sasse,   2004,   Whittaker   et   al.,   2010),   and   making   the  
conservative   estimate   that   these   will   remain   constant   over   our   digital   lifetimes,   we   will   actively   save  
around   125   thousand   documents,   115   thousand   emails   and   120   thousand   digital   photographs.   How  
 18  

people   organise   and   maintain   this   information   will   obviously   have   a   strong   bearing   on   their   success   in  
exploiting  that  information  in  the  future.    

Certain  types  of  management  also  take  place  more  often  than  we  might  expect.  For  certain  items  
such  as  files  and  emails,  people  are  perpetually  and  actively  engaged  in  re-­‐organisation,  as  reflected  by  
the   frequent   small   modifications   they   make   to   their   information.   For   example   a   longitudinal   study  
(Boardman   and   Sasse,   2004)   found   that   people   create   a   new   file   folder   every   three   days   and   they   make   a  
new  email  folder  every  5  days.  In  each  case,  the  new  structure  reflects  the  fact  that  people   are  constantly  
reflecting  on  how  their  information  is  currently  organised  and  finding  it  to  be  inadequate.  However,  as  we  
mentioned  in  the  keeping  section,  people  seldom  engage  in  major  reorganisations  or  extensive  deletion.  
Instead   they   tend   to   incrementally   modify   existing   structures.   They   are   highly   unlikely   however   to  
monitor  and  re-­‐organise  photos  or  contacts,  for  reasons  that  will  become  clear.      

People  also  make  management  mistakes.  They  often  engage  in  counterproductive  behaviours  in  
organising   their   information.   Studies   of   web   bookmarking   show   people   construct   complex   hierarchical  
bookmarking   systems   (Abrams   et   al.,   1998,   Aula,   2005).  Yet   we   have   already   seen   that   users   never   access  
42%  of  the  bookmarks  they  organise  for  later  retrieval  (Tauscher  and  Greenberg,  1997).  Efforts  organising  
emails  may  also  not  bear  fruit.  Email  filing  accounts  for  10%  of  total  time  in  email  (Bellotti  et  al.,  2005),  
and   yet   information   is   usually   accessed   by   browsing   the   inbox   or   search,   rather   than   folder   access  
(Whittaker,   2005,   Whittaker   et   al.,   2007,   Tang   et   al.,   2008).   With   personal   photos   they   may   make   the  
opposite  type  of  mistake  and  fail  to  organise  information  when  there  is  a  clear  need  to  do  so.  For  example  
a   study   of   personal   photo   retrieval   showed   a   failure   to   impose   even   rudimentary   organisation   -­‐   in   part  
because   people   believed   that   they   would   be   able   to   retrieve   their   photos   without   needing   to   organise  
them  (Whittaker  et  al.,  2010).  

Semantic  organisation  

Organising   information   is   a   fundamental   cognitive   activity.   One   basic   approach   is   to   apply  

conceptual   organisation   to   information.   Even   newborn   infants   categorise   objects,   with   natural  
psychological   categories   tending   to   be   based   around   exemplars   or   prototypes.   For   example,   people’s  
concept  of  ‘bird’  is  based  around  exemplars  such  as  robins,  rather  than  unusual  cases  such  as  penguins.  
Our   judgements   and   reasoning   are   influenced   by   the   extent   to   which   particular   instances   are   similar   to  
those  exemplars  (Rosch  et  al.,  1976,  Rosch,  1978).    

When   managing   personal   information,   there   are   two   different   and   separate   aspects   to  
organisation   that   are   important   for   effective   exploitation.   We   call   these   mental   and   physical   cueing.   As  
many   psychological   studies   have   shown,   the   mental   act   of   imposing   organisation   on   information   makes   it  
inherently   more   memorable.   Organising   things   within   a   consistent   conceptual   structure   means   that,   at  
recall,  one  item  may  mentally  trigger  memory  of  a  related  one,  so  applying  semantic  organisation  is  highly  
effective   in   promoting   recall   (Baddeley,   1995,   Craik   and   Lockhart,   1972).   Organisation   helps   recall,   even   if  
people   don’t   have   direct   access  to  their  organisational  scheme  at  retrieval.  For  example  in  a  recent  study  
we  showed  that  the  simple  act  of  organising  conversational  information  by  taking  notes  increased  recall  
even   when   people   didn’t   use   their   notes   at   retrieval   time   (Kalnikaite   and   Whittaker,   2008).   Organisation   is  
also  important  because  the  products  of  organisational  efforts  can  themselves  be  used  as  physical  retrieval  
cues.   Appropriate   notes   can   serve   as   cues   that   remind   us   about   information   items   that   we   might  
otherwise  have  forgotten  (Kalnikaite  and  Whittaker,  2007,  2008).  Well  chosen  folder  names  serve  to  cue  
 19  

people  about  their  contents  and  organisation  (Bergman  et  al.,  2003,  Lansdale,  1988,  Jones  and  Dumais,  
1986,  Jones  et  al.,  2005).    

Organisation  and  labelling  are  in  the  mainstays  of  most  computer  operating  systems.  The  main  
way   that   people   organise   their   digital   information   is   to   recursively   sort   it   into   categories   (in   directories,  
folders   or   subfolders)   and   then   apply   meaningful   labels   to   these   folders   and   subfolders.   The   act   of  
applying  organisation  may  help  retrieval  by  mental  cueing,  as  well  as  generating  a  navigable  conceptual  
structure   with   folder   labels   serving   as   physical   retrieval   cues.   Note   also   that   folders   usually   contain   a  
strong  spatial  component  –  with  subfolders  ‘sitting  inside’  super-­‐ordinate  items,  and  this  too  can  help  cue  
retrieval  (Jones  and  Dumais,  1986).    

Temporal  organisation  and  reminding  There  is  a  second,  less  obvious,  type  of  organisation  that  
has  been  less  extensively  researched.  We  have  already  seen  that  some  important  information  that  people  
deal  with  is  actionable.  Further  it  is  usually  the  case  that  those  actions  are  required  to  happen  by  a  certain  
time,   e.g.   to   meet   a   certain   deadline.   People   must   therefore   ensure   that   actionable   information   is  
organised  in  such  a  way  that  it  is  encountered  at  the  right  time,  allowing  the  deadline  to  be  met.  This  is  
the   problem   of   reminding.   It   is   no   good   having   an   extensive   organisational   structure   allowing   access   to  
any  item,  if  you  forget  the  deadline  relating  to  that  information.  Reminding  will  turn  out  to  be  a  critical  
problem  in  what  follows,  especially  in  the  case  of  email,  where  actionable  items  are  prevalent.    

Most   psychology   research   on   organisation   has   looked   at   natural   categories   (e.g.   how   we  
mentally  organise  places,  events,  names  and  faces).  But  it  has  not  looked  at  the  types  of  information  we  
are  addressing  here,  namely  synthetic,  human-­‐generated  information  such  as  documents,  emails,  photos  
or   web   pages.   Nevertheless   there   is   considerable   HCI   and   library   science   research   looking   into   people’s  
preferences   for   organising   such   personal   data.   For   example,   people   prefer   to   relocate   their   documents  
spatially   rather   than   using   keyword   search   (Barreau   and   Nardi   1995,   Bergman   et   al.,   2008).   This   spatial  
organisation   works   even   better   when   the   document   space   is   three-­‐dimensional   although   this   may   not  
scale  well  to  large  number  of  files  (Robertson  et  al.  1998).  However,  there  are  limits  to  the  utility  of  spatial  
organization:   semantic   labels   are   stronger   retrieval   cues   than   spatial   organization   alone,   although  
combinations  of  semantic  and  spatial  organization  can  enhance  performance  (Jones  and  Dumais,  1986).  
And  semantic  and  spatial  cues  are  enhanced  when  these  are  self-­‐selected,  rather  than  being  chosen  by  an  
external  party  (Bergman  et  al.,  2003,  Lansdale  and  Simpson,  1990).  There  is  also  evidence  for  the  utility  of  
temporal  organisation  as  a  retrieval  cue.  People  can  successfully  retrieve  documents  by  associating  them  
with   personal   or   public   events   that   happened   close   to   the   time   that   the   documents   were   encountered   or  
created  (Ringel  et  al.,  2003).  The  importance  of  temporal  factors  is  also  shown  by  log  files  of  search  tools  
revealing  a  bias  towards  retrieval  of  highly  recent  information  (Dumais  et  al.  2003;  Cutrell  et  al.  2006a).  

In  addition  to  these  overall  organisational  preferences,  other  work  has  explored  different  types  
of   management   strategy   and   what   motivates   people   to   choose   them.   We   now   describe   strategies   for  
paper,   digital   files,   email,   web   documents   and   photos.   We   review   the   types   of   management   strategies  
employed,  what  influences  people’s  choice  of  strategy  and  the  trade-­‐offs  between  strategies.    

Several   recent   papers   have   argued   that   manual   organisation   of   our   personal   data   will   soon  
become  obsolete.  Improvements  in  desktop  search  will  mean  that  documents  emails  and  web  pages  can  
be   easily   retrieved   without   the   need   for   active   organisation   (Russell   and   Lawrence,   2007,   Cutrell   et   al.,  
2006a).  This  is  an  appealing  idea.  We  have  seen  that  management  activities  are  onerous  and  difficult  for  
 20  

users,   who   may   invest   in   organisational   efforts   that   are   not   always   directly   successful.   We   will   discuss  
these  claims  in  more  detail  when  we  discuss  exploitation  techniques  and  evaluate  the  efficacy  of  different  
search  tools.    

4.2  MANAGING  PAPER  

Malone   (1983)   conducted   a   pioneering   study   into   people’s   organisational   habits   for   paper,  
identifying   two   main   strategies   he   called   filing   and   piling.   Filing   involves   constructing   an   exhaustive,  
hierarchical   taxonomy,   with   semantically   related   items   stored   in   each   subcategory.   In   contrast   piling   is  
more   laissez   faire,   usually   resulting   in   shallower,   less   systematic   hierarchies.   Piles   tend   to   be   fewer   in  
number   with   each   pile   containing   more   items,   with   looser  associations   between   items   stored   in   the   same  
pile.  Items  may  also  be  in  a  common  pile,  because  they  were  first  generated  or  acquired  at  the  same  time.  

There   are   clear   trade-­‐offs   between   these   two   organisational   strategies.   Piles   are   easier   to   create  
and   maintain,   as   they   are   less   systematic.   They   have   a   less   clear   organisational   structure   with   more   items  
in   each   pile,   which   may   make   retrieval   within   each   pile   more   inefficient.   But   because   there   tend   to   be  
fewer  piles  in  total,  this  leaves  fewer  potential  locations  to  be  searched  which  may  compensate  for  this  
lack   of   organisation.   Fewer   piles   may   also   mean   that   users   visit   each   pile   more   frequently   and   end   up  
being   more   familiar   with   the   contents   of   each.   Files,   in   contrast,   require   more   effort   at   creation   time   and  
more   maintenance.   However   they   offer   benefits   at   retrieval,   providing   a   more   coherent   retrieval  
structure  along  with  more  relevant  labels  as  cues.  These  advantages  may  be  offset  by  the  fact  that  there  
may  be  more  categories,  so  files  may  have  more  levels  to  navigate.  Files  may  also  fall  into  disrepair,  with  
too  many  levels/distinctions  being  too  infrequently  visited,  making  distinctions  between  categories  harder  
to  remember.    

In   the   move   study   described   above,   we   investigated   when   and   why   people   choose   filing   or   piling  
strategies.  The  distinction  between  filers  and  pilers  was  not  absolute,  instead  being  one  of  degree.  All  our  
respondents   filed   some   information,   but   kept   other   information   in   desktop   piles.   We   classified   users  
according   to   how   likely   they   were   to   file   information.   Based   on   the   predominant   strategies   that   people  
described  in  our  interviews,  we  identified  a  threshold  of  40%.    
 
Pilers   often   amass   information   without   attempting   to   systematically   organise   it.   This   laissez   faire  
approach  should  lead  to  an  accumulation  of  unscrutinised  information  before  the  office  move.  However  
we   found   to   our   surprise   that   pilers   had   smaller   original   archives.   They   also   had   less   preserved  
information  than  filers  after  cleaning  out  their  archives.  Why  then  did  filers  amass  more  information?  Our  
interviews  suggested  one  possible  reason  is  premature  filing:  filers  may  file  information,  which  turns  out  
to  be  of  little  utility,  that  they  later  have  to  discard.  If  filers  are  more  likely  to  incorporate  documents  of  
uncertain  quality  into  their  filing  systems,  we  might  expect  them  to  throw  away  more  reference  materials  
than   pilers   in   preparing   for   the   move.   This   was   not   true   for   all   documents,   but   was   true   for   reference  
documents.  

There  were  also  differences  between  strategies  in  terms  of  data  acquisition.  We  expected  pilers  
to  acquire  information  faster,  because  they  tend  not  to  scrutinize  incoming  data  as  carefully.  We  looked  
at   data   acquisition   rates,   in   separate   analyses   of   original   and   preserved   (i.e.   post-­‐move)   information  
volumes.   For   both   measures,   pilers   tended   to   be   slower   to   acquire   original   as   well   as   preserved  
information,  when  we  allow  for  the  number  of  years  they  had  been  in  the  company.    
 21  

Given   their   more   systematically   organised   systems,   we   expected   filers   to   find   it   easier   to   find  
data,  and  they  should  access  their  data  more  often.  Contrary  to  our  expectations,  pilers  had  accessed  a  
greater  percentage  of  documents  than  filers  in  the  last  year.  Why,  were  pilers  more  likely  to  access  recent  
data?   The   interviews   revealed   both   strategies   had   strengths   and   weaknesses.   With   a   piling   strategy,  
information   is   more   accessible:   it   can   be   potentially   located   in   a   relatively   small   number   of   piles   that  
people  frequently  sift  through.  The  result  of  this  is  that  valuable  frequently  accessed  information  moves  
to   the   top   of   the   piles,   and   less   relevant   material   ends   up   located   lower   down   the   pile.   This   pattern   of  
repeated   access   allows   people   to   identify   important   information,   discarding   unused   or   irrelevant  
information.  

But  the  lack  of  a  coherent  system  with  piling  has  some  disadvantages.  Taken  to  excess,  piles  can  
dominate   not   just   working   surfaces,   but   all   areas   of   the   office.   However,   even   though   filing   is   more  
systematic,   it   does   not   always   guarantee   easy   access   to   information.   With   complex   data,   filing   systems  
can  become  so  arcane  that  people  forget  the  categories  they  have  already  created,  leading  to  duplicate  
categories.  Accessing  only  one  of  these  duplicates  leads  to  incomplete  retrieval,  because  some  part  of  the  
original   information   will   be   neglected.   This   illustrates   a   general   disadvantage   to   filing   strategies:   they  
incur   a   large   overhead   for   constructing,   maintaining   and   rationalising   complex   organisations   of  
documents.  Similar  findings  are  reported  in  a  study  comparing  folders  and  tags  as  methods  of  organising  
personal  information  (Civan  et  al.,  2008).    

A  final  possible  reason  why  filers  access  proportionally  less  of  their  data  is  that  they  simply  have  
more  stuff.  There  are  finite  constraints  on  how  much  data  one  can  access.  Filers  have  more  data,  and  in  
consequence  they  are  able  to  access  less  of  it.  Consistent  with  this  is  the  fact  that  the  absolute  amounts  
of  data  accessed  by  both  groups  were  very  similar.    

We   also   expected   filers   to   be   quicker   to   rationalise   their   data   in   preparing   for   the   move,   given  
the  greater  care  they  have  taken  to  initially  organise  their  data.  But  there  were  no  differences  in  packing  
time  for  filers  and  pilers.  This  could  be  because  pilers’  greater  organization  is  offset  by  having  more  data  
to  sift  through.  And  contrary  to  our  predictions,  pilers  found  it  subjectively  easier  to  rationalise  archives  in  
preparation   for   the   move.   Why   was   this?   Despite   the   fact   that   filers   discarded   more   reference  
information,   they   generally   found   it   difficult   to   discard   filed   documents,   partly   because   of   the   investment  
they   had   already   made   in   managing   that   information.   Filers   therefore   seemed   less   disposed   to   discard  
information   they   had   invested   effort   in   organising.   In   contrast,   unfiled   information   seemed   easier   to  
discard.  

Finally   we   looked   at   what   determined   strategy   choice.   Although   job   type   influenced   strategy  
somewhat   (e.g.   secretaries   were   more   likely   to   be   filers),   in   general   strategy   seemed   to   be   more   affected  
by  dispositional  factors.    

4.3  MANAGING  DIGITAL  FILES  A ND  FOLDERS  

We  access  our  files  and  folders  on  a  daily  basis,  and  their  organisation  has  clear  importance  for  
our  everyday  digital  lives,  yet  there  have  been  relatively  few  studies  of  how  people  organise  their  digital  
files   and   what   affects   this   organisation.   One   exception   to   this   is   a   study   by   Boardman   and   Sasse   (2004)  
which   looked   at   the   structure   of   people’s   personal   data,   finding   that   on   average   people   had   57   folders  
where   the   average   folder   depth   was   3.3.   That   study   also   documented   different   filing   strategies,   finding  
that   58%   of   people   systematically   filed   information   items   when   they   created   them,   a   further   35%   left  
 22  

many  items  unfiled  (in  a  manner  similar  to  paper  piling),  with  a  small  proportion  (6%)  leaving  most  items  
unfiled.   In   some   cases,   people   did   not   file   actionable   documents   (i.e.   those   that   they   were   currently  
working  on),  instead  leaving  them  in  obvious  places  such  as  the  desktop  where  they  would  be  reminded  
about   them.   Boardman   and   Sasse   also   looked   at   the   types   of   folders   that   people   created,   identifying   2  
main   classes:   project   and   role   oriented.   Finally   they   looked   longer   term   to   see   whether   management  
strategies  changed  over  time  but  found  little  evidence  for  this.    

Two   other   studies   looked   at   the   structure   of   people’s   file   systems.   Gonçalves   &   Jorge   (2003)  
studied   the   folder   structure   of   11   computer   scientists   using   Windows   (8),   Linux   (2)   and   Solaris   OS   (1).  
Their  results  show  extremely  deep,  narrow  hierarchies.  The  average  directory  depth  was  found  to  be  8.45,  
with  an  average  branching  factor   (which  is  an  estimation  of  the  mean  number  of  subfolders  per  folder)  of  
1.84   indicating   a   deep   and   narrow   hierarchy   structure.   In   contrast,   a   larger   scale   study   by   Henderson   and  
Srinivasan   (2009)   looked   at   the   folder   structure   of   73   university   employees   using   Windows   OS.   The  
structures  they  found  were  much  shallower,  being  only  3.4  folders  deep  on  average.  Folders  tended  to  be  
broader   with   an   average   of   4.1   subfolders   per   folder,   for   non-­‐leaf   folders.   Both   studies   found   relatively  
small  numbers  of  files  per  folder:  13  for  (Gonçalves  &  Jorge,  2003)  and  11.1  for  (Henderson  &  Srinivasan,  
2009).    

In   another   study   probing   the   reasons   why   people   generate   specific   folder   structures,   Jones   et   al.  
(2005)  interviewed  people  about  the  nature  of  their  folder  systems.  They  discovered  that,  consistent  with  
physical   cueing,   many   folders   were   seen   as   plans,   i.e.   structures   that   people   used   to   organise   their   future  
work.  They  found  that  folders  represented  main  tasks  and  subtasks  of  ongoing  projects,  serving  to  remind  
people   about   aspects   of   their   work   activity   that   needed   to   be   executed.   People   used   also   various  
workarounds   to   make   various   types   of   information   more   salient,   e.g.   by   labelling   folders   ‘aacurrent’  
instead   of   ‘current’   to   ensure   that   this   information   was   more   obvious   when   browsing   an   alphabetically  
ordered  folder  list.    

Bergman  et  al.  (2003,  2009)  also  document  workarounds  within  folders,  to  make  individual  files  
and  folders  more  salient,  at  the  same  time  avoiding  the  need  to  delete  information.  They  describe  how  
people  create  subfolders  for  older,  less  relevant  information  and  label  these  ‘archive’  or  ‘old’  to  reduce  
clutter  and  make  relevant  working  items  more  visible  in  the  main  active  folder.    

Another   important   aspect   of   digital   file   organisation   is   the   adaptive   nature   of   active   folders.  
Bergman   et   al   (2008)   showed   that   the   most   common   strategy   for   accessing   personal   information   is  
navigation   through   the   folder   system,   with   this   type   of   access   occurring   many   times/day.   One   implication  
of   this   continual   re-­‐access   is   that   users   are   likely   to   discover   suboptimal   organisation,   leading   them   to  
adaptively   modify   their   file   and   folder   structures.   Adaptive   maintenance   and   modification   will   turn   out   to  
be  important  when  we  discuss  archives  that  are  much  less  frequently  accessed,  which  often  turn  out  to  be  
poorly  structured.  For  example,  people  add  an  average  of  5.9  new  files  to  their  work  collection  each  day,  
creating  a  new  file  folder  every  3  days.  In  contrast,  with  digital  pictures,  months  may  elapse  between  new  
folders   being   created,   with   negative   effects   of   people’s   ability   to   retrieve   those   pictures   (Whittaker   et   al.,  
2010).  

More  recently,  new  types  of  tool  have  been  developed  to  support  different  types  of  organisation.  
One  example  is  tagging.  Phlat  (Cutrell  et  al.,  2006b)  is  a  system  that  allows  users  to  apply  multiple  labels  
to  a  given  information  item,  rather  than  storing  it  in  a  single  folder  location.   Tagging  has  the  advantage  of  
providing  richer  retrieval  cues  (as  multiple  labels  are  available  as  retrieval  terms)  as  well  as  allowing  users  
 23  

to   filter   sets   of   retrieved   items   in   terms   of   their   tagged   properties   (e.g.   ‘pictures’   +   ‘personal’   returns   files  
with   those   tags).   In   contrast,   current   file   and   folder   systems   are   more   restricted   in   terms   of   the   ways   that  
data   can   be   accessed   and   navigated   to.   If   a   file   is   stored   in   the   ‘work2008’   folder,  unless   I   can   recall   or  
navigate   to   that   exact   folder   location,   I   will   be   unable   to   relocate   that   data.   But   despite   these   putative  
advantages,   in   a   long-­‐term   field   trial   users   made   very   little   overall   use   of   tagging,   averaging   only   one  
query  per  week  with  the  Phlat  system.  It  seemed  from  user  comments  that  the  costs  of  creating  tags  may  
have  been  too  high  to  generate  enough  of  the  tags  needed  to  support  flexible  search  and  filtering.  This  
led  people  to  use  the  system  more  like  a  standard  desktop  search  tool.  Another  study  compared  tagging  
and  foldering  again  failing  to  find  clear  benefits  for  tags  (Civan  et  al.,  2008).  In  the  next  section  we  discuss  
how  social  tagging  may  reduce  some  of  these  costs  of  creating  personal  tags.    

4.4  MANAGING  EMAIL  

Actionable  Items  

Managing   email   is   complex,   and   different   from   paper   or   standard   digital   files.   A   critical   aspect   of  
email  is  that  it  contains  many  actionable  messages.  To  be  effective,  people  need  to  organise  actionable  
information  in  such  a  way  that  they  are  reminded  of  what  they  need  to  do  when.  This  means  that  users  
have  to  organise  action  oriented  information  so  that  they  will  encounter  it  when  they  need  to  do  so.  We  
first  describe  how  users  process  actionable  messages.  We  then  turn  to  what  they  do  with  informational  
messages,  which  are  treated  more  like  paper  and  standard  digital  files.  

For  actionable  items,  deferral  is  inevitable.  Only  a  small  proportion  of  actionable  messages  can  
be  dealt  with  at  once,  and  most  actionable  items  must  wait  to  be  processed.  Dabbish  et  al.  (2005)  found  
that   on   average   37%   of   messages   that   require   replies   are   deferred,   which   equates   to   about   4   deferred  
messages   per   day.   If   people   forget   these   deferred   tasks,   this   can   create   major   headaches   both   for   the  
user  and  their  organisation.    

Whittaker   and   Sidner   (1996)   found   that   the   most   prevalent   strategy   for   reminding   about  
actionable  messages  is  to  leave  them  in  the  inbox.  Users  know  that  they  will  return  to  the  inbox  to  access  
incoming   unprocessed   messages   and   hopefully   be   reminded   about   their   outstanding   actionable  
messages.  Dabbish  et  al.  too  report  that  actionable  items  are  left  in  the  inbox  around  79%  of  time.  We  
called  this  strategy  ‘no  filing’.    

Whittaker   and   Sidner   also   showed   the   importance   of   using   the   inbox   to   prompt   visual   reminding  
by   observing   the   failure   of   other   strategies:   25%   users   had   experimented   with   a   strategy   of   filing  
actionable   items   in   a   "todo",   folder.   In   fully   95%   of   these   cases,   this   folder   was   abandoned,   because  
people  had  to  explicitly  remember  to  go  to  it,  open  it  and  review  its  contents.  This  extra  effort  contrasts  
with   being   reminded   about   outstanding   actions   by   the   mere   fact   of   seeing   them   in   the   inbox   when  
reading   new   email.   Although   other   studies   (Bellotti   et   al.,   2003)   suggest   that   some   users   change   their  
work   practices   to   exploit   ‘todo’   folders,   this   demands   extra   cognitive   steps.   Paradoxically,   these   users  
have   to   actively   remember   to   look   for   their   reminders.   In   contrast,   items   in   the   inbox   are   encountered  
naturally  as  a  side-­‐effect  of  accessing  new  messages.    

Of   course   there   are   also   disadvantages   to   leaving   actionable   items   in   the   inbox:   these   reminders  
may   be   difficult   to   spot   if   the   user   receives   many   new   messages.   Incoming   messages   visually   displace  
older  pending  actionable  items  -­‐  requiring  the  user  to  continually  scroll  through  their  inbox  to  ensure  that  
 24  

these  items  are  not  ‘out  of  sight  and  out  of  mind’  (Whittaker  et  al.,  2003,  Whittaker,  2005,  Whittaker  and  
Sidner,  1996).  Tang  et  al.  (2008)  looked  at  the  proportion  of  their  inbox  that  users  had  constantly  visible,  
finding  that  on  average  only  25%  of  inbox  emails  were  in  view.  The  remaining  75%  of  messages  were  not  
therefore   serving   as   direct   visual   reminders   for   outstanding   actions   –   compromising   their   ability   to  
remind.  

Other   users   try   to   keep   their   inbox   clear   by   filing   incoming   actionable   items   in   dedicated   task  
related   folders   (Bellotti   et   al.,   2005,   Whittaker   and   Sidner,   1996).   Whittaker   and   Sidner   dubbed   these  
people   ‘frequent   filers’   and   documented   how   25%   of   users   create   such   folders.   There   are   obvious  
advantages   of   this   strategy:   removing   such   items   from   the   inbox   keeps   the   inbox   trim   and   also   allows  
users  to  focus  better  on  new  and  important  information.  However  these  benefits  may  be  outweighed  by  
disadvantages:  users  are  required  to  create,  maintain  and  continually  check  these  task  folders.  Failure  to  
file  appropriately  can  also  have  severe  consequences,  if  they  file  important  information  and  forget  about  
it.  

A  final  strategy  for  actionable  items  is  a  hybrid  of  the  above.  Whittaker  and  Sidner  identified  a  
final  group  accounting  for  35%  of  their  users  who  engaged  in  ‘spring  cleaning’.  These  people  would  wait  
until   huge   amounts   of   information   accumulated   in   their   inboxes,   making   it   hard   to   identify   actionable  
items.   They   would   then   engage   in   extensive   filing   to   rationalise   their   inbox.   The   process   would   then   be  
repeated  with  the  inbox  gradually  growing  in  size  until  another  ‘crisis’  is  experienced  and  extensive  filing  
takes  place  once  more.    

What   determines   which   strategy   people   choose   when   processing   actionable   emails?   Whittaker  
and   Sidner   (1996)   looked   at   the   impact   on   strategy   choice,   of   organisational   role   and   incoming   volume   of  
messages.  Managers  were  more  likely  to  receive  greater  volumes  of  email,  but  there  was  no  evidence  of  a  
direct  relationship  between  strategy  and  role.  As  with  our  paper  study,  it  may  be  that  dispositional  factors  
are  an  important  determinant  of  strategy  choice.  This  is  supported  other  research  demonstrated  relations  
between  cognitive  style  and  strategy  (Gwizdka  2004a;  2004b).  

Other  studies  of  email  have  found   some  support  for  these  management  strategies  (Whittaker  et  
al.,   2002,   Whittaker,   2005,   Mackay,   1988,   Dabbish   et   al.,   2005,   Bellotti   et   al.,   2005,   Fisher   et   al.,   2008).  
However  later  work  indicates  few  instances  of  pure  ‘no  filers’,  i.e.  people  with  absolutely  no  folders  who  
are   totally   reliant   on   their   inboxes   for   task   management.   Balter   (2000)   both   extended   the   set   of  
management   strategies,   and   also   argued   that   people   move   sequentially   from   being   an   active   filer,   to  
spring   cleaner   and   later   no   filer,   as   the   volume   of   email   they   receive   increases.   He   argues   that   those  
receiving  the  highest  volumes  of  email  are  those  with  these  least  time  to  organise  it.    

Informational  Messages  

We  now  look  at  how  users  organise  informational  messages.  A  substantial  percentage  of  emails  
are  informational  as  opposed  to  actionable  (Dabbish  et  al.,  2005,  Whittaker  and  Sidner,  1996).  Users  also  
experience  problems  in  processing  informational  emails.  Observations  of  email  behaviour  show  that  users  
spend  huge  amounts  of  time  overall  in  organising  emails:  on  average  10%  of  people’s  total  time  in  email  is  
spent  filing  messages  (Bellotti  et  al.,  2005).    

Again   Whittaker   and   Sidner   (1996)   examined   why   users   have   problems   with   filing   such  
information.   There   are   several   reasons   why   creating   folders   for   informational   messages   is   hard.  
 25  

Generating  and  maintaining  folder  collections  requires  considerable  effort.  Filing  is  a  cognitively  difficult  
task  (Lansdale,  1988).  Just  as  with  the  keeping  decision,  successful  filing  is  highly  dependent  on  being  able  
to  envisage  future  retrieval  requirements.  It  is  hard  to  decide  which  existing  folder  is  appropriate,  or,  if  a  
new  folder  is  needed,  how  to  give  it  a  memorable  name  that  will  be  appropriate  for  the  retrieval  context  
in  which  it  will  be  needed.  

Again,  as  we  saw  in  the  keeping  section,  another  reason  for  not  filing  is  that  users  want  to  use  
the  deferral  strategy  and  postpone  their  judgment  about  the  value  of  information.  Users  do  not  want  to  
create  archives  containing  information  that  later  turns  out  to  be  useless  or  irrelevant.  They  are  aware  that  
creating  overly  complex  archives  may  make  it  harder  to  access  truly  valuable  information.    

Furthermore,   folders   may   not   be   useful   after   they   are   constructed.   Users   may   not   be   able   to  
remember  folder  labels,  especially  when  users  have  large  numbers  of  older  folders.  Research  combining  
multiple  studies  shows  that  people  have  an  average  of  around  39  email  folders  (Whittaker  et  al.,  2007).  
When   filing   they   therefore   have   to   remember   the   definition   of   each   and   to   be   careful   not   to   introduce  
duplication   by   creating   new   folders   that   are   synonymous   with   pre-­‐existing   ones.   Duplication   of   folders  
detracts  from  their  utility  at  retrieval.    

In   addition,   folders   can   be   too   small   to   be   useful.   A   major   aim   of   filing   is   to   coerce   the   huge  
number  of  undifferentiated  informational  inbox  items  into  a  relatively  small  set  of  folders  each  containing  
multiple   related   messages.   Filing   is   clearly   not   successful   if   the   number   of   messages   in   a   given   folder   is  
small.   If   a   folder   contains   only   one   or   two   items,   then   creating   it   has   not   significantly   reduced   the  
complexity  of  the  inbox,  nor  gathered  together  significant  amounts  of  related  material.  

Our   data   show   that   filing   often   fails:   on   average   35%   of   users'   folders   contain   only   one   or   two  
items.   Later   studies   duplicate   these   observations   although   finding   a   lower   percentage   (16%)   of   such  
‘failed’   folders   (Fisher   et   al.   2006).   Not   only   do   these   tiny   ‘failed   folders’   not   significantly   reduce   the  
complexity  of  the  inbox,  they  introduce  the  dual  overheads  of:  (a)  creating  folders  in  the  first  place,  and  
(b)  remembering  multiple  folder  definitions  every  time  there  is  a  decision  about  filing  a  new  inbox  item.  
This  cognitive  overhead  is  illustrated  by  the  fact  that  the  larger  the  number  of  folders  a  user  has,  the  more  
likely  that  person  is  to  generate  ‘failed  folders’  containing  only  one  or  two  items  (Whittaker  and  Sidner,  
1996).   Of   course   a   small   number   of   these   failed   folders   may   represent   new   activities   that   the   user   is  
planning   to   carry   out   (Bergman   et   al.,   2003,   Boardman   &   Sasse,   2004,   Jones   et   al.,   2005),   but   such  
planning  cannot  account  for  all  of  these  tiny  folders.    

Folders  can  also  fail  because  they  are  too  big.  When  there  are  too  many  messages  in  a  folder,  it  
becomes  unwieldy.  And  as  the  relationships  between  different  messages  within  the  folder  become  more  
tenuous,  the  benefit  of  keeping  them  together  is  much  reduced.  With  large  heterogenous  folders,  it  can  
be  extremely  difficult  to  collate  related  items,  or  find  a  target  item  (Whittaker  and  Sidner,  1996).  

Elsweiler  et  al  (2008)  looked  at  the  impact  of  filing  strategy  on  users’  memory  for  their  emails.  
Frequent   filers   tended   to   remember   less   about   their   emails.   This   result   is   consistent   with   our   earlier  
observations   about   premature   filing.   Filing   information   too   quickly   can   lead   to   the   creation   of   archives  
containing   spurious   information,   and   quick   filing  also   means   that   users   aren’t   exposed   to   the   information  
frequently  in  their  inbox,  making  it  hard  for  them  to  remember  its  properties  or  even  its  existence.    
 26  

Thus  email  users  experience  cognitive  difficulties  in  creating  folders  for  informational  messages.  
In  addition,  the  payoffs  for  this  effort  may  not  be  great:  folders  can  be  too  large,  too  small  or  they  may  be  
too   numerous   for   people   to   remember   individual   folder   definitions.   In   consequence,   folders   may   be   of  
restricted  use  either  for  retrieval  or  for  collating  related  messages.  As  we  have  seen,  some  users  finesse  
this   problem:   instead   of   filing   informational   messages,   they   simply   leave   them   all   in   their   inbox.   More  
recent   work   has   tried   to   support   this   strategy   by   introducing   new   techniques   such   as   thread   based  
viewers  which  we  describe  in  the  technology  trends  section  (6.1).    

4.5  MANAGING  WEB  PAGES  

Unlike   email,   web   information   is   largely   not   actionable:   users   may   want   to   ensure   that   they  
remember  to  read  a  webpage,  but  in  general  there  aren’t  negative  consequences  for  failing  to  do  this.    

One   prevalent   form   of   managing   web   information   is   to   bookmark   encountered   webpages.  There  
have   been   numerous   studies   looking   into   how   people   organise   their   bookmarks.   Two   early   studies  
documented  the  number  of  bookmarks  created  as  well  as  their  underlying  structure.  For  example,  Abrams  
et  al.  (1998)  found  that  6%  of  respondents  had  no  bookmarks,  10%  had  1-­‐10,  24%  had  11-­‐25,  44%  had  26-­‐
100,  14%  had  101-­‐300,  and  2%  had  300+  bookmarks.  And  Boardman  and  Sasse  (2004)  found  that  people  
organised   their   bookmarks   into   an   average   of   17   folders.   Another   study   (Bruce   et   al,   2004)   observed  
further   strategies   people   use   for   organising   useful   web   information   that   they   encounter.   In   addition   to  
bookmarking,   users   might   forward   themselves   a   link   in   email,   print   the   page,   copy   the   link   into   a  
document,  generate  a  sticky  note,  or  rely  on  memory.    

More  recent  work  with  more  modern  web  browsers  has  revisited  bookmarking.  Aula  et  al.  2005  
looked  at  people’s  bookmark  collections  finding  that  92%  have  bookmarks,  with  an  average  of  220  links,  
although  there  is  huge  variance  in  collections:  21%  of  people  have  fewer  than  50  bookmarks,  and  6%  have  
none.   The   largest   collection   contained   2589   links   with   425   folders.   Most   of   Aula   et   al.’s   (2005)   informants  
reported   major   problems   in   organising   and   managing   their   collections.   Consistent   with   other   studies  
(Tauscher   and   Greenberg,   1997)   users   often   bookmarked   information   that   they   never   subsequently  
revisited.  In  contrast,  other  studies  showed  that  users  are  unwilling  to  create  new  bookmarks  fearing  that  
creating  bookmarks  for  information  of  unclear  utility  will  clutter  their  existing  set  of  useful  bookmarks  -­‐  
compromising  the  utility  of  useful  items  (Aula  et  al.,  2005,  Wen,  2003).  Aula  et  al  also  found  that  the  key  
for  success  with  complex  bookmark  collections  is  the  extent  to  which  users  actively  exploit  and  maintain  
their  collection  of  links.  There  was  a  subgroup  of  heavy  users  of  bookmarks,  who  had  collections  of  over  
500  links.  These  heavy  users  tended  (like  email  spring  cleaners)  to  clean  up  their  collections  from  time  to  
time   deleting   unused   or   no   longer   functioning   links.   They   also   carefully   organised   bookmarks   into  
hierarchical  levels  (similar  to  a  file  system).  For  these  users  who  invested  organisational  effort,  bookmarks  
seemed   to   be   an   indispensable   tool.   Abrams   et   al.   also   looked   at   the   types   of   strategies   that   people   used  
for   organising   their   bookmarks.   They   found   4   main   types:   about   50%   of   people   were   sporadic   filers,   a  
further  26%  never  organised  bookmarks  into  files,  around  23%  created  folders  when  they  accessed  a  web  
page  and  around  7%  created  folders  at  the  end  of  a  session.  Creating  folders  also  seem  to  be  a  response  
to  having  too  many  bookmarks  on  a  drop  down  list,  so  that  people  with  fewer  than  35  bookmarks  have  no  
folders  but,  beyond  this  threshold,  folders  grow  linearly  with  the  number  of  bookmarks.  
   
Some   of   the   disadvantages   of   bookmarking   relate   to   the   costs   of   creating   and   maintaining  
collections,   especially   as   information   needs   change.   Recent   social   tagging   systems   such   as   Deli.cio.us,  
 27  

Dogear,   Onomi,   and   Citeulike,   may   finesse   some   of   these   problems.   These   social   tagging   systems   allow  
users  to  create  multiple  labels  for  the  same  data  potentially  providing  richer  retrieval  cues  (Cutrell,  2006b,  
Lansdale,  1988).  More  importantly  they  allow  tags  to  be  shared  between  users,  reducing  the  cost  of  tag  
creation  for  each  user.  Of  course  the  approach  raises  important  questions.  Do  different  users  agree  on  a  
common  classification  of  information,  or  do  they  generate  inconsistent,  orthogonal  tag  sets?  Numerous  
studies   have   shown   that   given   sufficient   numbers   of   users,   tag   sets   tend   to   stabilise   on   common  
descriptions   of   web   resources   so   that   people   can   exploit   others’   tags   (Golder   and   Huberman   2006,   Millen  
et   al.,   2007).   Furthermore,   with   suitable   user   interface   design,   e.g.   text   completion,   problems   such   as  
inconsistent  spellings  can  be  finessed,  as  well  as  promoting  greater  awareness  of  others’  tags  (Millen  et  
al.,  2007).  If  enough  people  are  prepared  to  tag,  social  tagging  seems  a  useful  tool  that  removes  some  of  
the  costs  associated  with  standard  individual  bookmarking  methods.    

4.6  MANAGING  PHOTOS  

Photos   are   very   different   from   emails   and   web   pages,   tending   to   be   self-­‐generated   (like   many  
files),   and   being   neither   informational   nor   actionable.   They   are   also   perceived   to   be   highly   important   and  
often   irreplaceable   (Petrelli   et   al.,   2008,   Whittaker   et   al.,   2010).   How   then   do   people   organise   them?  
Recent  studies  show  that  people  manage  to  organize  photos  using  rather  rudimentary  structures  (Kirk  et  
al.,  2007,  Whittaker  et  al.,  2010).    

Whittaker   et   al.   (2010)   looked   how   parents   organised   family   photo   archives.   They   found   that  
these  collections  tended  to  have  very  little  hierarchical  structure,  and  were  organised  more  like  piles  than  
files.   Participants   typically   relied   on   a   single   main   picture   storage   location   (such   as   the   “My   Pictures”  
folder).   For   participants   with   multiple   storage   devices   (computers   and   hard   drives)   there   was   usually   a  
single  main  storage  folder  for  each  device.  People  usually  stored  their  pictures  in  that  location  in  a  single  
level  flat  hierarchy  with  minimal  subfolders.  Furthermore,  when  a  target  folder  was  opened  and  scanned  
that   folder   often   contained   heterogeneous   data,   containing   pictures   that   relate   to   multiple   events  
(possibly  because  they  were  uploaded  at  the  same  time  and  never  subsequently  reorganised).    

How  can  we  explain  this  lack  of  organisation?  Previous  work  has  highlighted  how  participants  are  
able  to  exploit  their  familiarity  with  recently  taken  pictures  to  quickly  scan,  sort  and  organize  materials  for  
sharing  with  others  (Kirk  et  al.,  2007).  Possibly  as  a  result  of  these  experiences  with  recent  pictures,  our  
participants  may  have  expected  themselves  to  be  very  familiar  with  their  entire  picture  collection,  and  as  
a   result   weren’t   motivated   to   organize   their   collections   carefully.   In   most   cases,   it   seemed   that   people  
hadn’t   accessed   the   vast   majority   of   their   pictures   since   they   were   uploaded.   We   saw   evidence   of   this  
when   participants   retrieved   pictures.   Photos   almost   always   appeared   in   the   “list”   view.   However,  
participants   universally   preferred   to   view   pictures   in   the   thumbnail   view   for   easier   scanning.   Had   the  
participants  previously  opened  these  folders,  the  thumbnail  view  would  have  remained  at  the  interview.  
And   because   participants   seldom   accessed   pictures,   they   didn’t   discover   how   poorly   organised   these  
were.   One   reason   for   the   lack   of   organisation   and   unfamiliarity   is   that   parents   typically   have   very   little  
spare  time  to  organize  their  photos.  One  participant  commented  that  his  attitude  to  photos  was  “collect  
now  –  organize  later  –  view  in  the  future”.    

Another   potential   way   to   organise   might   be   to   annotate   pictures.   However,   consistent   with  
earlier   studies   (Frohlich   et   al.,   2002,   Kirk   et   al.,   2007,   Rodden   and   Wood,   2003),   we   found   very   little  
evidence  of  annotation.  One  reason  for  failing  to  annotate  is  that  this  is  onerous.  Another  problem,  also  
 28  

observed  in  earlier  studies  (Kirk  et  al.,  2007,  Rodden  and  Wood,  2003),  is  that  users  may  fail  to  annotate  
because   they   are   unaware   that   they   are   likely   to   forget   key   aspects   of   pictures.   People   can   currently  
remember  detailed  information  about  recent  pictures  and  this  may  mean  they  have  little  motivation  to  
annotate  pictures  for  the  eventuality  that  they  will  forget.    

4.7  MANAGEMENT  SUMMARY  

1. Management  is  a  difficult  activity  for  users,  because  it  requires  people  to  predict  when  or  how  
information   will   be   accessed.   To   create   effective   organisation   users   have   to   anticipate   the  
context  in  which  they  will  be  accessing  information.  And  for  action  oriented  items,  they  have  to  
anticipate  exactly  when  they  will  need  those  items.    
2. Information   properties   have   a   major   impact   on   management   strategy:   actionable   items   often  
require  deferral,  so  people  need  to  be  reminded  about  them.  Various  tracking  strategies  facilitate  
reminding,   including   leaving   actionable   information   in   one’s   workspace,   as   well   as   using  
dedicated  task  folders.  There  are  trade-­‐offs  between  these  strategies:  keeping  information  in  a  
workspace  affords  constant  reminding,  but  it  reduces  efficiency  as  that  workspace  can  become  
cluttered  with  many  unrelated  actionable  items.  And  one  specific  problem  with  using  the  email  
inbox   for   reminding   is   that   as   new   items   arrive   they   tend   to   displace   older   actionable   items  
leading  them  to  be  ‘out  of  sight  and  out  of  mind’.  The  disadvantage  of  dedicated  task  folders  is  
that  these  need  to  be  constantly  accessed  and  monitored.    
3. For   informational   items,   people   use   two   main   strategies,   filing   and   piling.   There   are   surprising  
advantages   for   a   paper   piling   strategy.   Pilers   manage   to   build   up   smaller   archives,   with   more  
frequent   access   to   information   in   their   archive.   In   addition,   we   found   problems   with   filing  
including,   premature   filing   of   low   value   information   leading   people   to   generate   complex  
collections  of  information  that  are  of  little  utility.    
4. For   informational   items,   users   experience   difficulty   in   categorising   information,   failing   to  
accurately  predict  the  context  in  which  they  will  want  to  retrieve  that  information.  People  create  
folders   that   are   both   ‘too   big’   –   containing   large   collections   of   heterogeneous   items   and   ‘too  
small’   containing   one   or   two   items   in   a   folder   that   is   seldom   used.   People   can   also   create  
duplicate  folders  for  the  same  content.  All  this  makes  filing  error-­‐prone.    
5. Both   users’   dispositions   and   the   volume   of   information   they   receive   may   influence   the   type   of  
organisational   strategy   they   use.   Users   who   receive   large   volumes   of   incoming   information   are  
under  pressure  to  keep  their  workspaces  clear  (otherwise  they  may  overlook  important  deferred  
actionable   items)   but   they   are   the   people   who   are   least   likely   to   have   the   time   to   file   and  
organise  their  information.    
6. Certain   types   of   information   such   as   web   pages   and   photos   are   infrequently   re-­‐accessed.  
Infrequent   access   may   mean   that   people   fail   to   realise   what   information   they   have   available   and  
how   poorly   organised   it   is.   Tags   don’t   seem   to   be   useful   in   the   context   of   personal   files,   but   they  
do   seem   to   have   benefits   in   a   web/intranet   context   where   people   can   reduce   the   cost   of  
annotation  by  sharing  others’  labels.  

5. EXPLOITATION  

5.1  OVERVIEW ,  PROBLEMS  AND  STRATEGIES  

In   this   section,   we   first   contrast   exploitation   with   classic   information   seeking   and   foraging  
behaviours,   go   on   to   describe   different   strategies   for   exploitation,   as   well   as   the   costs   and   benefits   of  
these  strategies.  

Exploitation  Not  Information  Seeking.    

 29  

Exploitation   is   different   from   information   foraging   and   classic   information   seeking.   In   both  
foraging   (Pirolli   et   al..,   1999,   2007)   and   classic   information   seeking   (Belkin,   1980,   Marchionini,   1995,  
Wilson,  1999),  the  target  information  is  seen  as  being  totally  new.  Exploitation  is  different  in  several  ways.  
First,  retrieval  structures  are  usually  self-­‐  rather  than  publicly  generated  (Lansdale,  1988,  Bergman  et  al.,  
2003).  In  other  words,  people  are  searching  their  own  organisation  and  not  a  public  database.  Second  the  
exploiter   may   remember   significant   details   about   the   target   information   item   and   how   it   has   been  
organised.    

For   example,   Goncalves   and   Jorge   (2004)   asked   participants   to   tell   stories   about   3   personal  
documents   they   had   recently   worked   on.   People   could   remember   a   great   deal   about   these   documents  
with  the  most  salient  characteristics  being  age,  location  and  purpose  of  the  document.  Blanc-­‐Brude  and  
Scalpin   (2007)   also   found   that   location,   format,   age,   keywords   and   associated   events   were   frequently  
remembered.  Because  people  remember  this  information,  access  is  not  purely  reliant  on   external  publicly  
provided  metadata  (‘scent’  in  the  terminology  of  information  foraging).  Instead  it  is  mediated  by  cueing:  
where   cues   can   be   mental   (the   internal   cognitive   information   users   remember   about   the   target   before  
they  begin  to  access  it)  or  physical  (external  triggers  provided  by  well-­‐chosen  folder  or  file  names  as  users  
carry  out  their  search).  Indeed  as  we  saw  earlier,  management  activities  have  the  predominant  purpose  of  
constructing  personal  organisations  that  promote  future  exploitation.    

Exploitation   therefore   involves   reconstruction   of   partially   familiar   personally   organised  

information,  rather  than  evaluation  of  unfamiliar  publically  organised  data.  A  further  difference  concerns  
success  criteria:  while  information  seeking,  it  is  often  enough  to  access  information  that  satisfies  certain  
general  properties  (‘cheap  flights  to  Spain’),  where  multiple  documents  may  satisfy  this  search.  In  contrast  
when   accessing   personal   information,   the   user   often   has   a   specific   document   in   mind   -­‐   making   the  
criterion   for   success   much   more   stringent.   Of   course   such   prior   knowledge   may   make   retrieval   easier.  
During   access,   users   may   quickly   recognise   the   target   document,   so   they   do   not   have   to   scrutinise   it   to  
determine  its  relevance  as  they  would  an  unknown  web  page.  But  in  other  ways  access  to   very  specific  
information  can  be  made  harder  when  access  is  only  satisfied  if  a  specific  item  is  found,  and  there  may  be  
strong  feelings  of  frustration  about  failure  to  locate  that  item  (Whittaker  et  al.,  2010).    

Exploitation  Strategies  

Exploitation   success   depends   on   the   match   between   cues/structures   generated   for   future  
retrieval  and  the  extent  to  which  they  match  that  future  retrieval  context.  Note  that  even  if  people  rely  on  
search,  they  still  have  to  generate  the  relevant  search  terms  to  guarantee  success,  and  this  requires  them  
to   reconstruct   important   aspects   of   the   target   document   (e.g.   title,   keywords,   date).   If   there   is   a   good  
match   between   organisational   cues   and   the   retrieval   context,   then   retrieval   will   succeed.   But   to   create  
effective   retrieval   cues,   users   need   to   successfully   anticipate   when   and   how   they   will   consume  
information.    

There  are  4  main  ways  that  we  access  personal  information.    

One  very  straightforward  way  to  access  information  is  to  navigate  for  it.  For  information  items  
such   as   files,   we   navigate   within   self-­‐generated   hierarchies   of   folders   and   subfolders   to   locate   our  
information.  People  usually  manually  traverse  their  organizational  hierarchy.  They  visually  and  recursively  
scan  within  each  folder  (either  actively  by  sorting  the  items  by  attribute  or  by  using  the  system  default),  
until  they  locate  the  folder  that  contains  the  target  item.  
 30  

Search   is   another   way   to   access   personal   information.   An   important   emerging   technology   for  
exploitation   is   desktop   search,   allowing   users   to   locate   information   from   within   their   own   file   systems,  
using   key   word   queries,   in   the   same   way   they   conduct   web   searches.   Users   first   generate   a   query   by  
specifying   some   property   of   the   target   item,   including   at   least   one   word   related   to   the   name   of   the  
information  item,  and/or  the  text  that  it  contains  (full  text  search)  and/or  any  metadata  attribute  relating  
to  that  item  (e.g.  the  date  the  item  was  created).  The  desktop  search  engine  then  returns  a  set  of  results  
from   which   the   user   selects   the   relevant   item.   Search   has   elsewhere   been   characterised   as   a   form   of  
teleporting   whereby   users   move   directly   to   the   target   information,   without   the   intermediate   steps   that  
characterise  navigation  (Teevan  et  al.,  2004).    

A   third   access   method,   orienteering   is   a   hybrid   combining   both   navigation   and   search   (Teevan   et  
al.,  2004).  When  orienteering  users  may  generate  a  search  query  to  locate  a  particular  resource  page  or  
folder   and   then   manually   navigate   to   the   target   or   they   might   begin   by   accessing   a   link,   and   use  
information  from  that  link  to  generate  a  new  search  query.    

Finally,  new  technologies  such  as  tagging  allow  users  to  apply  multiple  labels  to  an  information  
item  both  on  the  desktop  (Cutrell  et  al.,  2006a)  or  on  the  web  (deli.cio.us,  Flickr).  They  allow  users  more  
flexibility   in   how   they   categorise   the   item   (as   more   than   one   label   can   be   applied.   Multiple   tags   mean  
richer  retrieval  cues,  as  the  same  information  can  be  accessed  via  several  different  tags.    

The  above  strategies  apply  to  personal  information.  When  people  incorporate  public  information  
into   their   personal   schemes   (e.g.   web   bookmarking,   or   history   lists)   more   varied   strategies   are   possible  
(Aula   et   al.,   2005,   Bruce   et   al.,   2004,   Jones   et   al.,   2003,   Obendorf   et   al.,   2007).   For   example   users   can  
deliberately  bookmark  valued  information  or  save  it  to  disk  and  then  navigate  back  to  this  data.  Or  they  
can  apply  less  effortful  strategies  such  as  accessing  information  via  the  history  list  (a  list  of  sites  visited),  
or  use  the  browsers  ‘back’  button  to  reaccess  recent  information.    

Costs  and  Benefits  of  Exploitation  Strategies  

If  the  fit  between  the  organisation  that  users  construct  and  the  retrieval  context   is  inexact,  even  
careful   management   strategies   may   not   guarantee   successful   retrieval.   The   wrong   classification   of  
information  can  ‘hide  it’  from  the  user,  reducing  the  chance  of  quick  retrieval  (Kidd,  1994;  Malone,  1983;  
Whittaker  and  Sidner,  1996).  Putting  information  in  a  folder  may  decrease  its  ability  to  remind  which  may  
be   vital   for   actionable   information.   In   addition,   because   categorisation   is   itself   cognitively   challenging,  
users   may   create   spurious   folders   that   are   seldom   accessed,   and   which   may   make   classification   of   new  
information  harder  (Fisher  et  al.,  2006,  Whittaker  and  Sidner,  1996).    
 
What  then  are  the  trade-­‐offs  between  navigation  and  search  for  accessing  personal  information  
items?   There   are   clear   benefits   to   navigation.   Accessing   information   using   a   personally   constructed  
organisational   hierarchy   is   predictable   and   includes   a   spatial   component   which   users   find   valuable  
(Barreau  and  Nardi,  1995,  Bergman  et  al.,  2008,  Jones  and  Dumais,  1988,  Robertson  et  al.,  1988).  Access  
takes  place  in  incremental  stages,  so  that  users  obtain  rapid  feedback  about  the  progress  of  their  access  
efforts,  being  able  to  backtrack  if  they  find  they  have  accessed  the  wrong  branch  of  their  file  hierarchy.  At  
the  same  time,  there  are  disadvantages  to  navigation,  compared  with  search.  In  complex  organisational  
structures,  navigation  can  be  inefficient,  and  taking  a  wrong  step  early  in  the  access  process  may  require  
extensive   backtracking   depending   on   the   precise   nature   of   the   organisation   scheme   (Hearst,   1999).  
 31  

Furthermore,   users   have   to   remember   at   retrieval   time,   how   information   was   classified,   which   can   be  
difficult  when  there  are  multiple  categorisation  possibilities  (Lansdale,  1988,  Russell  and  Lawrence,  1997).    

There   are   also   potential   advantages   of   search   when   accessing   personal   information.   Search   does  
not   depend   on   users   remembering   the   exact   storage   location   or   precisely   how   they   classified   their  
information;  instead,  they  can  specify  in  their  query  any  attribute  they  happen  to  remember  (date,  name,  
filetype)  (Lansdale,  1988).  Search  may  also  be  more  efficient:  user  can  potentially  retrieve  information  in  
one  step,  via  a  single  query,  instead  of  using  multiple  operations  to  navigate  to  the  relevant  part  of  their  
folder   hierarchy.   More   radically,   search   also   potentially   finesses   the  management   problem,   as   users   don’t  
have  to  apply  organisational  strategies  that  exhaustively  anticipate  their  future  retrieval  requirements.    

The   same   dichotomy   between   navigation   and   search   does   not   apply   to   actionable   items.   Here  
very   different   strategies   must   be   used.   Reminding   is   key,   so   that   information   must   be   organised   in   such   a  
way   that   users   encounter   it   opportunistically.   Neither   search   nor   navigation   through   complex   file  
organisations   are   appropriate   support   for   actionable   items,   as   both   require   deliberate   acts   to   seek   out  
data,   whereas   the   primary   characteristic   of   actionable   items   are   that   these   should   trigger   automatic  
reminding.   This   is   clearly   a   very   hard   problem:   effective   reminding   means   users   don’t   just   want   to   re-­‐
encounter   actionable   information,   they   want   to   see   it   exactly   when   or   where   they   need   it.   Actionable  
information   presented   at   the   wrong   time   may   be   highly   distracting,   and   it   turns   out   that   very   different  
strategies  are  needed  for  actionable  than  informational  items.  

Turning   now   to   public   data   (such   as   web   data)   that   people   want   to   incorporate   into   their  
personal   organisational   schemes,   it   is   apparent   that   users   may   have   less   incentive   to   manage   public   data,  
because   this   is   less   highly   valued,   being   less   personally   relevant   or   unique   (Boardman   et   al.,   2003,  
Whittaker   and   Hirschberg,   2001).   There   are   also   clear   trade-­‐offs   between   different   exploitation   strategies  
for   public   data   (Bruce   et   al.,   2004).   Although   browsers   now   offer   support   in   the   form   of   ‘suggestions’,  
regenerating  prior  searches  still  requires  considerable  effort  in  remembering  search  terms,  especially  as  
search  is  often  iterative  involving  multiple  searches  relating  to  a  specific  information  need,  some  of  which  
may  result  in  ‘deadends’  (Morris  et  al.,  2008).  Retracing  successful  navigation  is  also  hard.  Users  have  to  
remember  which  links  they  traversed.  Bookmarking  requires  people  to  remember  which  information  they  
have   bookmarked,   as   well   to   maintain   bookmarking   collections.   And   more   passive   strategies,   (e.g.   relying  
on  the  history  list)  means  that  users  have  to  navigate  through  poorly  structured  traces  of  every  piece  of  
information   they   accessed   rather   than   just   information   that   they   thought   was   valuable   (Wen,   2003,  
Morris  et  al.,  2008).  In  all  cases,  retrieval  may  be  made  more  difficult  by  the  changing  nature  of  the  web,  
which  may  alter  the  content  of  pages  users  previously  accessed.    

We   now   discuss   different   strategies   that   people   choose   for   exploitation   of   different   types   of  
information:  namely  files,  emails,  photos  and  web  information.    

5.2  ACCESSING  FILES  

There   have   been   significant   recent   developments   in   desktop   search.   One   limit   of   older   search  
engines,   such   as   those   provided   as   part   of   the   Windows   and   Macintosh   operating   systems,   is   that   they  
allow   users   to   search   only   one   data   format   at   a   time.   Following   the   Stuff   I’ve   Seen   (SIS)   initiative   (Dumais  
et   al.,   2003),   newer   search   engines   support   multiple   formats   –   files,   emails,   instant   messages   and   Web  
history   can   be   accessed   within   the   same   search   query.   They   therefore   potentially   address   the   ‘project  
 32  

fragmentation’  problem  -­‐  where  information  items  related  to  the  same  project  are  automatically  stored  in  
different  locations  often  because  they  depend  on  different  applications  (Bergman  et  al.,  2003,  Dragunov  
et  al.,  2005).  Modern  search  engines  are  also  substantially  faster  than  older  ones,  with  more  sophisticated  
interfaces   to   specify   their   search   choices   (Farina,   2005;   Lowe,   2006).   Search   is   now   also   incremental,  
returning  results  as  soon  as  the  user  begins  typing  their  query.  This  incrementality  allows  users  to  refine  
their  query  in  light  of  the  results  returned,  and  truncate  the  query  after  typing  just  a  few  characters  if  the  
target  item  is  already  in  view.    

In  a  recent  study,  (Bergman  et  al  2008),  we  investigated  whether  advanced  desktop  search  was  
replacing  navigation  as  the  main  method  for  file  access.  We  used  multiple  different  methods  (longitudinal  
evaluation,  large  scale  cross  sectional  surveys),  as  well  as  examining  different  search  engines  (Windows  XP  
search,   Google   desktop,   Mac   Spotlight,   Mac   Sherlock).   Users   reported   how   often   they   searched   versus  
navigated,   to   their   files.   We   verified   the   accuracy   of   this   self   report   data   by   collecting   logfiles   which  
allowed   us   to   correlate   self   report   data   with   actual   behavioural   access   logs.   Self   reports   were   very  
accurate  and  highly  correlated  with  actual  behaviour,  with  statistical  correlations  being  around  0.94.    

We  know  that  organisation  requires  effort  -­‐  having  to  create  and  maintain  appropriate  structures  
that  anticipate  retrieval,  as  well  as  having  to  remember  those  structures  during  exploitation.  Given  these  
new  search  engine  capabilities,  we  expected  users  to  shift  away  from  relying  on  navigation  for  file  access  
and   become   increasingly   reliant   on   desktop   search.   We   expected   that   people   having   access   to   desktop  
search  engines  with  advanced  features  would  be  more  likely  to  access  their  files  using  search  than  those  
who  were  using  older  search  engines  without  those  features.    

Contrary   to   our   expectations,   we   found   that   navigation   was   still   users’   preferred   method   for  
accessing   their   files.   First,   regardless   of   search   engine   properties,   there   was   a   strong   overall   navigation  
preference:  users  estimated  that  they  used  navigation  for  56-­‐69%  of  file  retrieval  events  and  searched  for  
only   4-­‐16%   of   events.   The   remaining   accesses   were   when   users   relied   on   shortcuts   or   used   recent   files   to  
access  what  they  had  been  working  on.  Further,  the  effect  of  improving  the  quality  of  the  search  engine  
on  search  usage  was  limited  and  inconsistent.  Although  Google  Desktop  (which  was  fast,  incremental,  and  
supported  cross  format  search)  led  to  more  usage  than  Windows  XP  search,  there  was  no  evidence  that  
other  more  advanced  features  induced  greater  usage.  For  example,  both  Mac  search  engines  were  used  
equally   often,   despite   the   fact   that   the   later   version,   Spotlight,   was   faster,   as   well   as   supporting   cross  
format,  incremental  search.  Similar  results  using  very  different  qualitative  methods  have   also  shown  that  
pure  search  is  uncommon.  Instead  users  often  combine  search  with  navigation  (Teevan  et  al.,  2004).    

How   can   we   explain   why   retrieval   strategy   seemed   to   be   largely   independent   of   search  engine  
quality?   One   reason   is   that   search   often   seemed   to   be   used   as   a   last   resort   when   users   could   not  
remember   a   file’s   location.   Bergman   et   al.,   (2008)   asked   users   to   characterise   exactly   when   they   used  
search   as   opposed   to   navigation   and   found   that   between   83-­‐96%   of   the   times   when   people   searched,  
they  did  so  because  they  were  unable  to  remember  the  files’  location.  When  they  can  remember  they  rely  
on  navigation.    

It   also   seems   that   in   the   majority   of   cases   users   can   remember   where   files   are   located.   This   is  
unsurprising   if   we   think   that   for   common   tasks,   we   are   frequently   accessing   and   modifying   information  
related   to   specific,   often   recent,   items   (Dumais   et   al.,   2003),   and   this   reinforces   our   memory   for   those  
items   and   their   locations.   And   as   we   have   seen,   people   are   able   to   remember   substantial   amounts   of  
information  about  recent  files  (Blanc-­‐Brude  and  Scalpin,  2007,  Goncalves  and  Jorge,  2004).  The  conclusion  
that   search   is   used   only   when   people   can’t   remember   the   location   of   a   file,   is   supported   by   other   studies.  
 33  

Jones,  et  al  (2005)  found  that  only  7%  of  users  were  happy  with  the  idea  that  they  could  dispense  with  
folders  even  when  desktop  search  was  available.    

5.3  ACCESSING  EMAIL  

Accessing  information  in  email  is  a  critical  problem,  given  the  amount  of  time  that  people  spend  
processing  it  and  the  fact  that  it  is  both  a  ‘todo’  list  for  actionable  information  as  well  as  an  archive  for  
more  informational  data  (Duchenaut  and  Bellotti,  Whittaker  and  Sidner,  1996,  Whittaker,  2005).  

A  critical  aspect  of  email  management  is  to  ensure  that  actionable  items  are  dealt  with  to  meet  
specific  commitments.  The  previous  section  documented  that  the  most  common  reminding  strategy  is  to  
leave  such  items  in  the  email  inbox,  hoping  that  these  will  be  re-­‐encountered  on  returning  to  the  inbox  to  
process   new   incoming   information   (Bolter,   2000,   Bellotti   et   al.,   2005,   Dabbish   et   al.,   2005,  Mackay,   1988,  
Whittaker  and  Sidner,  1996,  Whittaker,  2005,  Whittaker  et  al.,  2007).  Variants  of  the  ‘inbox  as  todo  list’  
strategy  include  altering  the  status  of  actionable  items  that  have  been  read  resetting  the  status  of  such  
messages  so  that  they  appear  to  be  unread  and  hence  bold  in  a  standard  browser  (Whittaker,  2005).    

Despite   the   central   role   of   email   in   everyday   work,   we   know   relatively   little   about   how   people  
actually  retrieve  information  from  email.  One  exception  is  a  study  by  Elsweiler  et  al.  (2008)  who  looked  at  
people’s   ability   to   remember   emails.   Participants   were   usually   able   to   remember   whether   or   not   a  
message  was  in  their  collection.  Also  memory  for  specific  information  about  each  message  was  generally  
good   with   users   often   remembering   multiple   attributes.   People   remembered   content,   purpose   or   task  
related   information   best,   correctly   recalling   over   80%   of   this   type   of   information   –   even   when   items   were  
months   old.   They   were   less   good   at   remembering   sender   information,   and   memory   for   this   type   of  
information  tended  to  decay  rather  quickly.  Memory  for  temporal  information  was  worst  of  all,  dropping  
to  around  50%  correct  over  several  months.  In  all  cases,  memory  was  affected  by  both  the  age  and  size  of  
the   email   archive,   with   users   remembering   less   when   they   had   bigger   archives   or   when   they   were  
required  to  remember  older  items.    

Dumais   et   al   (2003)   also   examined   email   access   in   Stuff   I’ve   Seen   (SIS).   SIS   is   a   cross   format  
search  engine  allowing  users  to  access  files,  emails,  web  pages  by  issuing  a  query  in  a  single  interface.  It  
also   supports   results   sorting   via   attributes   such   as   date   or   author.   The   majority   of   searches   (74%)   were  
focused   on   email   as   opposed   to   files.   This   may   be   because   as   we   saw   earlier   (Bergman   et   al.,   2008),   if  
people  want  to  access  files,  they  do  so  using  navigation  rather  than  search.  When  searching  for  emails,  
there  was  a  very  strong  focus  on  recent  items,  with  21%  of  searched  for  items  being  from  the  last  week,  
and   almost   50%   from   the   last   month.   Many   of   these   searches   (25%)   included   the   name   of   the   email  
sender  in  the  query,  suggesting  that  (contrary  to  Elsweiler  et  al.,  2008)  that  sender  name  is  useful  retrieval  
cue   for   emails.   Elsewhere   we   exploit   the   salience   of   sender   name   in   the   ContactMap   system   which  
provides   a   specific   informational   view   on   email   data,   centred   around   network   models   of   sender   data  
(Whittaker  et  al.,  2004).  How  can  we  explain  the  prevalence  of  name  based  search  observed  by  Dumais  et  
al.,   when   compared   with   Elsweiler   et   al.’s   (2008)   results?   Part   of   the   difference   may   be   due   to   the   fact  
that  Dumais  et  al.  (2003)  observed  naturalistic  behaviours  which  tended  to  be  focused  around  retrieving  
recent   emails.   In   contrast   Elsweiler   et   al.   looked   at   longer-­‐term   access,   for   more   structured   lab   based  
tasks.  In  addition,  Dumais  et  al.  did  not  look  at  the  success  of  searches;  it  may  be  that  although  sender  
information  was  used  frequently  in  searches,  these  sender  searches  were  often  unsuccessful.    
 34  

5.4  ACCESSING  PHOTOS  

We   have   already   described   how   people   organise   their   digital   pictures   and   the   rudimentary  
management  strategies  that  they  employ.  As  with  email  research,  there  has  been  more  focus  on   photo  
management   and   rather   less   examining   exploitation.   Digital   photos   are   an   extremely   highly   valued  
resource   (Petrelli   et   al.,   2008,   Whittaker   et   al.,   2010),   so   we   should   expect   people   to   create   organisations  
ensuring   they   are   effective   at   accessing   these.   Indeed,   work   on   accessing   recently   taken   photos   shows  
that  people  are  good  at  retrieving  these  (Frohlich  et  al.,  2002).  Kirk  et  al  (2007)  asked  participants  to  sort  
recent  pictures  in  preparation  for  sharing  these  with  friends  or  family,  and  found  that  participants  were  
effective  in  finding  and  organising  pictures  taken  within  the  last  year.    

These  findings  contrast  with  our  own  work  where  we  looked  at  parents’  ability  to  retrieve  slightly  
older   family   pictures   (taken   more   than   a   year   ago).  Despite   the   fact   that   pictures   were   judged   as   being  
highly  valued,  participants  were  often  unsuccessful  in  accessing  such  older  pictures.    

We  asked  participants  to  name  significant  family  events  from  more  than  a  year  ago  that  they  had  
photographed   digitally.   In   a   subsequent   retrieval   task,   participants   were   asked   to   show   the   interviewer  
digital   pictures   from   3-­‐5   of   these   salient   past   events   concerning   their   children.   To   avoid   having   the  
participants   choose   events   that   they   could   easily   retrieve,   participants   weren’t   told   about   the   retrieval  
task   during   the   initial   interview.   The   interviewer   asked   participants   to   sit   at   their   computer   and   show   him  
pictures  relating  to  these  key  events.    

In   contrast   to   their   expectations,   our   participants   were   successful   in   retrieving   pictures   in   only  
slightly  more  than  half  of  the  retrieval  tasks  (61%).  In  the  remainder  (39%),  participants  simply  could  not  
find   pictures   of   significant   family   events.   Of   the   28   unsuccessful   retrieval   tasks,   21   (75%)   were   pictures  
that   the   participants   believed   to   be   stored   on   their   computer   (or   on   CDs)   but   which   they   subsequently  
could   not   find.   The   remaining   7   were   pictures   participants   initially   thought   were   stored   digitally,   but  
during  the  retrieval  process  changed  their  minds  into  thinking  were  taken  with  an  analog  camera.  

Based   on   participants’   comments   and   behaviour   during   and   after   search,   we   identified   several  
potential   reasons   for   their   unexpectedly   poor   retrieval   performance:   too   many   pictures,   distributed  
storage,   unsystematic   organisation,   false   familiarity,   and   lack   of   maintenance.   In   our   discussion   of  
management  we  have  already  talked  about  the  absence  of  systematic  organisation  and  the  tendency  to  
collect  too  many  pictures,  we  now  explore  the  implications  of  these  for  retrieval.    

The  most  frequent  explanation  participants  gave  for  their  retrieval  difficulties  was  that  they  had  
very  large  numbers  of  pictures  to  search.  Consistent  with  previous  work  (Frohlich  et  al.,  2002,  Kirk  et  al.,  
2007,  Rodden  and  Wood,  2003),  participants  felt  that  they  were  taking  many  more  digital  pictures  than  
they  had  with  analog  equipment.  All  participants  pointed  to  the  low  cost  of  capturing  large  numbers  of  
digital  pictures.  However,  during  retrieval  they  realized  that  having  too  many  pictures  has  its  price  when  
this  mass  of  pictures  competed  for  their  attention,  making  it  hard  to  locate  specific  ones.   Average  archive  
size  was  4475  pictures  but  with  huge  amounts  of  variation  (SD  3039).  This  is  a  striking  finding,  because,  
consistent   with   other   research   (Kirk   et   al.,   2005),   participants   all   made   definite   efforts   to   reduce   their  
overall   number   of   pictures.   For   example   they   deleted   around   17%   of   poorly   focused   or   unwanted  
pictures,  both  when  pictures  were  first  taken,  as  well  as  at  upload.    
 35  

Some  participants  attempted  to  account  for  their  poor  retrieval  by  arguing  that  they  hadn’t  given  
folders   meaningful   names.   However   67%   of   participants   made   efforts   to   apply   meaningful   labels   rather  
than   relying   on   software   defaults.   But   this   did   not   seem   to   guarantee   they   could   find   their   pictures,  
possibly  because  as  we  saw  in  the  management  section,  naming  schemes  were  inconsistent.  People  who  
used   meaningful   labels   were   neither   more   successful,   nor   faster   at   retrieving   pictures.   Participants’  
comments  and  behaviours  also  suggested  that  the  meaning  of  such  names  was  sometimes  forgotten  over  
time.   Finally,   participants   commented   on   difficulties   in   remembering   changes   over   the   years   in  
organisational  schemes  they  had  imposed  or  software  they  had  used.    

The   lack   of   organisation   in   people’s   collections   meant   that   they   were   over-­‐reliant   on   trial   and  
error  strategies  for  accessing  their  photos.  Consistent  with  studies  of  autobiographical  memory  (Brewer,  
1988;   Wagenaar,   1986),   some   of   our   18   participants   tried   to   use   knowledge   of   related   events   to  
remember   the   approximate   date   when   the   target   event   occurred   and   then   navigate   using   date  
information   to   the   folders   they   thought   might   contain   these   pictures.   Specific   folders   were   chosen  
because   their   name   (if   there   was   a   meaningful   name)  was   thought   to   relate   to   the   target   or   because   a  
folder  date  was  close  to  the  guessed  date.    

Others  tried  to  remember  the  exact  date  when  the  event  had  occurred  and  to  find  folders  from  
that   date.   This   worked   when   folders   had   been   labelled   with   correct   dates,   although   in   many   cases,   folder  
labels  were  purely  textual.  We  have  already  noted  problems  with  this  strategy.  First  participants  may  be  
unable   to   accurately   remember   the   date   of   the   target   event.   Second   the   date   label   itself   may   be  
inaccurate,   either   because   of   problems   with   camera   settings,   or   the   folder   date   represents   the   upload  
date  -­‐  as  opposed  to  when  the  picture  was  actually  taken.    

Overall  the  retrieval  strategy  used  most  often  seemed  to  resemble  trial  and  error:  users  would  
cycle   through   their   entire   photo   collection   accessing   folders   to   see   whether   they   contained   promising  
pictures,  moving  on  to  other  folders  if  they  did  not.  

5.5  ACCESSING  WEB  DOCUMENTS  

Accessing   web   pages   is   a   problem   that   has   been   much   studied.   Most   people’s   intuitions   about  
web  accesses  are  that  these  follow  the  pattern  of  foraging:  i.e.  that  people  predominantly  seek  out  new  
information   from   the   web   that   they   then   consume   for   the   first   time.   The   same   intuitions   also   lead   people  
to  think  that  the  typical  way  that  people  access  web  information  is  to  rely  on  search.    

One  possible  reason  for  this  belief  in  the  dominance  of  search  is  that  historically  web  tools  have  
moved  from  relying  on  navigation  via  human-­‐generated  categories  to  being  search  based.  Early  web  tools  
such   as   Yahoo!   provided   human-­‐generated   taxonomies   of   the   then   relatively   small   collection   of   web  
documents,   supporting   access   by   allowing   users   to   navigate   through   these   hierarchies.   However,   one  
limitation  of  these  manual  taxonomic  techniques  is  that  they  are  completely  impractical  for  the  billions  of  
documents  that  are  now  estimated  to  be  on  the  web.  Self-­‐report  studies  also  suggest  that  usage  of  web  
navigation  is  now  much  less  frequent,  with  people  reporting  a  far  greater  reliance  on  search  for  foraging  
(Kobayashi,  and  Takeda,  2000).    

In   reality,   however,   it   turns   out   that   search   is   less   frequent   than   we   might   expect.   Instead   of  
foraging   for   new   information,   users   tend   to   re-­‐access   previously   visited   data   using   a   variety   of   simple  
 36  

browser   techniques   including   following   links,   retyping   the   URL   or   exploiting   the   ‘back   button’   (Aula   et   al.,  
2005,  Bruce  et  al.,  2004,  Obendorf  et  al.,  2007).    

Many   studies   have   attempted   to   document   the   extent   to   which   web   accesses   involve  
information   seeking   versus   refinding   by   analysing   logfiles   and   history   lists.   Early   work   looking   at   students’  
browsing  behaviours  showed  that  a  characteristic  web  access  pattern  involved  ‘hub  and  spoke’  accesses,  
in  which  users  would  find  a  useful  authoritative  resource  –  a  ‘hub’.  They  would  then  navigate  out  to  the  
various   links   from   this   page   (‘spokes’)   usually   traversing   no   more   than   two   links   before   reaccessing   the  
hub   using   the   ‘back’   button   (Catledge   and   Pitkow,   1994).   Tauscher   and   Greenberg   (1997)   instrumented  
browsers  and  looked  at  the  rate  at  which  people  made  visits  to  previously  visited  sites.  They  documented  
a  recurrence  rate  of  58%,  finding  also  that  the  majority  of  overall  accesses  targeted  a  small  set  of  websites  
that  the  user  frequently  re-­‐accessed.  Revisits  are  prevalent,  as  indicated  by  the  use  of  the  ‘back’  button  
which   accounts   for   around   30%   of   web   actions.   In   addition   Tauscher   and   Greenberg   (1997)   found   that  
people  were  much  more  likely  to  reaccess  sites  that  they  had  been  to  recently.  Cockburn  and  Greenberg  
(2000)  carried  out  a  similar  study  finding  a  much  higher  frequency  of  accesses  (81%),  were  revisits.  

Another  study  conducted  by  Wen  (2003)  was  unusual  in  looking  at  the  success  of  refinding.  He  
asked   users   to   conduct   typical   web   access   sessions   and   then   subsequently   requested   them   to   retrieve  
information   that   they   had   found   useful   in   that   search   session.   Users   were   only   able   to   successfully  
reaccess   20%   of   the   sites   they   had   visited.   These   users   often   failed   to   bookmark   useful   information  
believing  that  doing  so  would  create  ‘clutter’  and  compromise  their  existing  bookmark  collections.  Finally,  
consistent  with  other  results  (Teevan  et  al.,  2004),  Wen  found  that  the  general  strategy  for  reaccess  was  
to  try  and  retrace  prior  actions,  rather  than  attempting  to  search  or  type  in  prior  URLs.  

Aula   et   al   (2005)   looked   at   users   self-­‐reported   strategies   for   web   search   and   reaccess.   They  
found  that  having  multiple  windows  or  tabs  open  was  very  common  because  reaccess  was  prevalent.  In  
addition,  the  most  commonly  reported  ways  to  reaccess  information  were  to:  re-­‐access  links,  search  for  it  
again,  directly  type  the  URL  or  to  save  pages  as  local  files.  This  confirms  the  results  of  an  observational  
study  by  Bruce  et  al.,  (2004)  that  documented  that  the  most  prevalent  strategy  for  refinding  was  to  type  
in  the  URL.  Other  access  strategies  were  much  less  prevalent,  e.g.  emailing  links  to  oneself,  adding  URLs  
to   a   website   or   writing   down   queries.   Finally   there  is   very   little   use   of   history   lists   for   reaccess.   Aula   et   al.  
found   various   problems   with   history   lists:   not   only   are   page   titles   often   misleading,   the   list   shows  
important  and  unimportant  results  intermingled  -­‐  making  it  hard  for  users  to  focus  on  valued  information.  
Both  Aula  et  al.,  (2005)  and  Wen  (2003)  also  noted  user  problems  with  re-­‐access:  one  problem  with   using  
search   to   exploit   information   that   it   is   an   iterative   process   often   involving   multiple   queries.   Users   may   try  
multiple   routes   to   finding   information   exploring   sites   that   later   turn   out   to   be   ‘deadends’.   In   trying   to  
recover   from   these   deadends,   users   often   couldn’t   regenerate   previous   accesses   that   had   been   more  
successful.  Users  also  couldn’t  recall  the  exact  method  that  they  had  used  for  access,  in  consequence  they  
had  problems  in  ‘reconstructing’  search  queries  for  information  that  they  had  originally  browsed  for.    

In   perhaps   the   best   controlled   study   of   revisiting,   Obendorf   et   al,   (2007)   preprocessed   sets   of  
URLs  for  25  users  finding  that  revisiting  rates  in  prior  studies  might  have  been  artificially  inflated  by  sites  
that   automatically   refreshed   without   user   intervention.   When   such   automatic   refreshes   are   controlled  
for,   they   found   revisitation   levels   were   around   41%.   They   also   documented   a   variety   of   general   strategies  
used  to  access  pages.  The  most  common  strategies  were:  using  a  hyperlink  (44%  of  accesses),  using  forms  
 37  

-­‐  including  the  use  of  search  engines  (15%),  ‘back’  button  (14%),  opening  a  new  tab/window  (11%),  typing  
in  the  URL  directly  (9%).  

Turning  specifically  to  revisits  (as  opposed  to  all  searches),  Obendorf  et  al.  again  found  that  the  
most  common  strategy  for  refinding  information  was  to  follow  links  (50%),  with  the  ‘back’  button  being  
the   next   most   common   strategy   (around   31%   of   time).   The   remaining   ‘direct   access’   strategies   (using  
bookmarks,  homepage  links,  history,  direct  entry  of  URL)  accounted  for  the  final  13%  of  accesses.  As  in  
previous  studies,  re-­‐accesses  tended  to  be  for  recently  accessed  sites:  73%  of  revisits  occur  within  an  hour  
of   the   first   visit,   which   makes   the   use   of   the   ‘back’   button   appear   rather   low.   One   possible   reason   for   the  
relatively   low   numbers   of   ‘back’   accesses   may   be   that   the   tabbing   facilities   provided   by   new   browsers  
mean  that  users  aren’t  as  reliant  on  ‘hub  and  spoke’  type  reaccesses.  They  can  therefore  keep  the  context  
of  their  ‘hub’  page  while  using  tabs  to  manage  follow-­‐up  ‘spoke’  pages.    

Finally  Obendorf  et  al.  looked  at  how  access  strategies  varied  as  a  function  of  the  length  of  time  
since  the  original  page  access.  Again  there  were  huge  recency  effects,  50%  of  revisits  occurred  within  3  
mins.   and   the   dominant   strategy   here   was   to   use   the   back   button,   presumably   because   the   target  
information   was   readily   available   in   the   browser   cache.   For   revisits   occurring   within   the   hour,   the   back  
button  and  links  were  the  most  common  ways  to  refind  data.  Between  an  hour  and  a  day,  back  button  
usage  hugely  decreased,  with  users  becoming  more  reliant  on  links  and  direct  access  (typing  in  the  URL).  
Between  a  day  and  a  week,  links  and  typing  URLs  were  the  most  common  strategies,  and  at  intervals  of  
greater   than   a   week,   use   of   links   dominated.   This   greater   reliance   on   link   usage   may   reflect   an  
orienteering   strategy   (Teevan   et   al.,   2004)   in   which   users   generate   plausible   sets   of   links   and   then   choose  
between  these  for  the  final  stage  of  access.  In  any  case,  the  results  clearly  show  that  access  strategies  are  
quite  varied  and  are  heavily  dependent  on  the  time  interval  between  initial  access  and  reaccess.  Part  of  
the   reason   for   this   is   technical:   for   very   short   term   re-­‐accesses   information   is   directly   available   in   the  
cache,   whereas   at   longer   intervals   this   is   unlikely   to   be   true.   In   addition,   there   are   cognitive   factors   at  
work  here.  At  medium  and  longer  reaccess  intervals  users  may  have  generated  several  windows  or  tabs,  
so  they  are  unable  to  remember  which  of  these  they  first  used  to  access  their  data.    

Finally  the  majority  of  revisits  (73%)  occur  within  an  hour,  12%  between  an  hour  and  a  day,  9%  
between  a  day  and  a  week  and  8%  at  longer  intervals.  As  we  have  seen,  the  time  between  accesses  is  a  
critical  factor  influencing  retrieval,  and  the  fact  that  the  majority  of  revisits  are  really  short  term  means  
that  certain  strategies  (such  as  using  the  back  button  or  link  based  access)  are  prevalent  overall.    

To   summarise,   then,   web   retrieval   often   involves   re-­‐accessing   previously   accessed   data.   Using  
links,  tabs  and  the  back  button  are  prevalent  for  more  recently  accessed  pages.  Search  tends  not  to  occur  
very   often.   Users   also   tend   to   access   a   small   number   of   sites   and   other   research   shows   that   familiarity  
also  influences  retrieval  strategy  (Capra  and  Perez-­‐Quinones,  2005).    

5.6  EXPLOITATION  SUMMARY  

1. During  exploitation,  people’s  preference  is  to  use  manual  methods  (folder  navigation/following  
links),  whether  this  is  for  regular  files  or  web  data.  Search  is  a  dispreferred  option  even  for  web  
documents.    
2. Search  is  not  successful  with  personal  photos  (content  based  techniques  are  weak,  and  there  is  
very  little  metadata),  people  therefore  have  to  rely  on  browsing  which  turns  out  to  be  ineffective  
in  many  cases  for  older  data.    
 38  

3. Emails  are  different  from  files:  search  can  be  useful  for  informational  items  because  people  are  
able   to   remember   certain   information   about   messages   (names/content),   at   least   in   the   short  
term.  However  reminding  is  needed  for  actionable  items,  and  search  can’t  be  used  because  it  is  a  
deliberate   act   that   implies   the   user   has   already   remembered.   Users   therefore   have   to   rely   on  
scanning   their   inboxes,   which   is   often   inefficient   because   of   the   amount   of   heterogeneous  
information  they  currently  contain.  
4. Web   Pages   –   despite   people’s   intuitions,   search   is   not   the   prevalent   way   to   access   web   data.  
Reaccesses   are   very   common   with   people   using   the   back   button   or   hyperlinks   as   their   main  
reaccess   methods.   Reaccesses   are   usually   for   recently   accessed   information   and   the   re-­‐access  
strategy  depends  on  how  recently  the  target  item  was  last  accessed.  
5. There  is  sometimes  a  mismatch  between  retrieval  structures  and  their  exploitation.  For  Photos,  
there   seems   to   be   a   failure   to   create   retrieval   appropriate   structures,   which   occurs   in   part  
because   these   are   not   frequently   accessed,   which   means   retrieval   is   unsuccessful   for   older  
materials.  For  Emails,  people  spend  large  amounts  of  time  creating  folder  structures  which  may  
not  always  be  exploited.  For  Web  documents,  people  often  create  structures  (such  as  bookmark  
collections)  that  aren’t  used  because  there  are  less  costly  ways  to  access  information.  They  also  
fail  to  create  structures  that  are  useful.  
6. Retrieval  has  clear  regularities  –  there  is  a  strong  bias  towards  access  of  recent  items,  as  well  as  a  
bias  towards  accessing  a  small  number  of  items  very  frequently.    

6. FUTURE  RESEARCH  
  What  then  are  pressing  future  issues  for  research  into  information  curation?  In  particular  since  
technology   is   so   important   in   this   area,   what   impact   will   emerging   technologies   have   on   keeping,  
management  and  exploitation?  

6.1  TECHNOLOGY  TRENDS  

Keeping  

Storage   is   now   so   cheap   that   we   no   longer   need   to   delete   items   because   they   are   consuming  
valuable   space.   One   general   shift   will   therefore   be   away   from   models   where   users   delete   information,  
either   when   it   is   first   encountered   or   during   later   ‘cleanups’.   Instead   people   will   tend   towards   ‘keeping  
everything’   (Jones,   2004,   Marshall,   2008a,b),   but   with   interfaces   that   provide   views   onto   what   is  
important  and  valuable  in  that  data.    

There   are   clear   advantages   to   this   ‘keep   everything’   approach.   We   know   that   users   find   deletion  
cognitively  and  emotionally  difficult,  and  they  are  also  concerned  that  they  will  end  up  deleting  valuable  
information   (Bergman   et   al.,   2009).   ‘Keeping   everything’   means   that   these   difficult   decisions   can   be   at  
least  partially  avoided,  although  the  consequence  is  that  we  need  new  approaches  to  management  and  
exploitation   if   users   aren’t   to   be   overwhelmed   by   kept   data.   In   this   spirit,   we   have   begun   to   build   user  
interfaces  that  keep  more  data  (assuaging  worries  about  deleting  valuable  information),  but  that  privilege  
information   that   is   valuable   or   important.   For   example,   motivated   by   a   study   of   users’   current  
workarounds   with   files   and   folders,   we   built   GrayArea   (Bergman   et   al.,   2009),   which   implements   a   two  
tier  view  of  each  folder,  with  the  main  view  showing  critical  documents.  The  secondary  area  (GrayArea)  is  
for  less  important  files  which  are  made  less  visually  salient,  but  still  potential  available.  A  user  evaluation  
showed  the  utility  of  this  interface  compared  with  the  standard  Windows  Explorer  method  of  managing  
files.  Of  course  one  problem  with  this  approach  is  that  it  requires  manual  organisation  to  generate  two  
 39  

tier  views,  and  we  are  exploring  (semi-­‐)  automatic  methods  for  learning  distinctions  between  these  two  
types  of  information,  in  an  attempt  to  reduce  the  burdens  of  manual  organisation.    

Other   technical   possibilities   involve   the   direct   application   of   machine   learning   to   address   the  
keeping   decision.   Indices   and   profiles   could   be   built   based   on   the   structure   and   content   of   people’s  
current   email,   files   and   web   documents.   These   could   also   include   information   about   which   information   is  
accessed  and  changed  most  frequently.  The  data  could  then  be  used  to  generate  an  ‘interest  profile’  for  
the  user  which  could  then  be  applied  to  incoming  emails  or  recently  accessed  web  pages.  If  for  example,  
an   incoming   email   or   viewed   webpage   bears   a   close   match   to   information   that   is   already   in   the   user’s   file  
system,  then  this  email  would  be  a  clear  candidate  for  keeping.  In  contrast,  an  email  bearing  no  relation  
to  the  user’s  interests  is  a  good  candidate  for  deletion.  One  problem  with  this  approach,  however  is  that  it  
might   be   very   effective   at   recognising   positive   candidates   for   keeping   but   rather   less   good   for   deciding  
what   should   be   rejected.   There   are   various   problems   with   automatically   deleting   information   that   is  
unrelated   to   the   user’s   current   profile.   Just   because   incoming   information   is   unrelated   to   the   user’s  
current   activity,   doesn’t   mean   that   it   is   irrelevant.   Unrelated   messages,   files   or   documents   might   just  
represent  an  exciting  new  opportunity,  an  emerging  new  area  or  a  potentially  important  new  contact,  and  
should  not  therefore  be  deleted.    

Management  

There   is   a   long   history   of   programs   being   built   to   support   management   (see   Whittaker   et   al.,  
2007   for   a   review),   in   particular   in   email   where   many   systems   try   to   file   or   filter   incoming   emails  
automatically  or  semi-­‐automatically.  There  are  various  problems  with  this  approach  however.  

One   critical   problem   is   that   users   fundamentally   don’t   trust   machine   learning   programs   (Pazzani,  
2000).  People  are  concerned  that  important  incoming  messages  might  be  misfiled.  It  is  clear  that  despite  
large  improvements  in  machine  learning  helped  by  the  existence  of  new  corpora,  that  programs  are  still  
errorful  (Whittaker  et  al.,  2002b,  Whittaker  et  al.,  2007).  And  while  programs  promise  to  correctly  classify  
documents  into  folders  with  relatively  low  error  rates,  we  still  lack  vital  empirical  data  about  what  error  
rates   are   acceptable   to   users.   Until   we   clearly   know   whether   users   will   at   best   tolerate   5%   of   misfiling  
then  we  don’t  know  what  quality  our  machine  learning  algorithms  need  to  be.    

One  response  to  the  errors  problem  is  to  use  semi-­‐automatic  methods.  Here  the  system  suggests  
to   the   user   where   a   document   might   be   filed,   and   the   user   is   asked   to   confirm   or   correct   this.   This  
approach  is  well-­‐liked  by  machine  learning  advocates,  because  it  provides  a  way  to  generate  structured  
feedback   on   the   algorithm   by   the   user   (Whittaker   et   al.,   2004,   Whittaker   et   al.,   2007).   But   there   is   a  
downside  to  this.  Unless  the  interface  is  well  designed,  so  that  suggestions  and  user  feedback  are  handled  
in   a   lightweight   manner,   then   the   effort   of   correcting   user   suggestions   may   be   greater   than   manual   filing.  
Feedback   and   suggestions   need   to   be   extremely   subtle   with   good   defaults,   otherwise   the   purported  
solution  (automatic  filing)  may  require  more  effort  than  users’  current  manual  filing  practices.    

Another,   perhaps   more   promising   approach   might   be   to   use   public   resources   to   organise  
personal  data.  For  example  systems  such  as  Phlat  (Cutrell  et  al.,  2006b)  and  Dogear  (Millen  et  al.,  2007)  
use  social  tags  as  ways  to  organise   personal  resources.  For  example,  a  document  in  my  filing  system  may  
inherit   tags   that   others   have   applied   to   that   document   in   a   public   archive.   This   approach   has   the   benefits  
that  user  generated  tags  are  often  more  appropriate  than  machine  generated  ones,  and  it  also  reduces  
the   management   costs   to   the   individual   user   who   has   access   to   rich   tags   without   generating   them  
 40  

him/herself.   However   there   are   various   unanswered   questions   here,   such   as   how   to   weight   the  
importance  of  personally  generated  versus  social  tags.  In  addition,  as  we  have  seen,  many  of  the  user’s  
most  important  documents  are  unique,  making  it  unlikely  that  public  tags  are  available  to  describe  them.    

Yet   another   approach   to   automatic   management   is   to   analyse   user   activity   to   determine  

importance   of,   and   relatedness   between,   documents.   A   common   intuition   is   that   documents   that   we  
access   frequently   are   more   likely   to   be   important,   as   are   recently   accessed   documents.   The   ‘my   recent  
documents’   shortcut   in   MS   Windows   capitalises   on   the   latter   intuition,   and   more   principled   algorithms  
have   also   been   built   to   capture   more   systematic   aspects   of   recency   (Tang   et   al.,   2007).   Other   systems  
have   used   social   information   to   profile   documents,   so   that   resources   that   are   frequently   accessed   by  
others  are  visually  privileged  over  those  that  are  less  frequently  accessed  (Kalnikaite  et  al.,  2008a).  

One  specific  area  where  machine  learning  might  be  extremely  beneficial  is  for  actionable  items,  
which   are   often   user’s   greatest   concern   when   processing   emails.   Work   on   analysing   email   content   has  
been   relatively   successful   in   predicting   whether   a   given   message   requires   a   response   (Cohen,   1996).  
Annotating   emails   with   this   information   and   presenting   this   in   the   interface   might   be   very   useful   in  
helping   people   keep   track   of   todos.   Another   approach   to   this   problem   is   thread   detection   and  
visualisation  which  is  now  a  part  of  newer  email  clients  (e.g.  Gmail),  and  research  prototypes  (Bellotti  et  
al.,   2003,   Tang   et   al.,   2008,   Venolia   and   Neustaedter,   2003,   Wattenberg   et   al.,   2005)   and   more   recent  
products   such   as   Gmail.   These   thread   viewers   attempt   to   reduce   inbox   ‘clutter’   by   clustering   related  
messages.   This   has   the   joint   benefits   of   collating   related   information   as   well   as   reducing   visual   distraction  
in  the  inbox.  Although  there  have  been  two  small  scale  evaluations  of  this  technique  (Bellotti  et  al  2003,  
Tang  et  al.,  2008),  as  yet  we  know  little  about  how  effective  these  techniques  might  prove  to  be,  although  
one  study  (Tang  et  al.,  2008)  suggests  that  threading  may  interfere  with  established  foldering  practices.    

Another  specific  area  where  we  can  expect  developments  in  curation  is  with  photos,  where  we  
have   seen   that   users   have   major   problems   with   management   and   exploitation   (Whittaker   et   al.,   2010).  
Standard  metadata  such  as  time  and  location  might  be  supplemented  with  GPS  and  compass  data  about  
where   a   camera   is   pointing   (allowing   inferences   about   what   the   shot   might   contain   as   well   as   content  
based   tagging).   GPS   data   might   also   indicate   where   a   photo   was   taken   (Kalnikaite   et   al.,   2010).   And  
specific   content   based   techniques   such   as   face   recognition   might   allow   familiar   people   to   be   tagged   in  
pictures,  a  tool  already  available  in  Picasa  and  on  the  Macintosh.  However  the  promise  of  face  recognition  
needs   to   be   evaluated   in   the   light   of   practical   concerns.   Nametags   may   be   most   important   for  
infrequently  encountered  people  whose  identity  the  user  is  likely  to  forget:  but  will  users  be  prepared  to  
tag   large   numbers   of   people   and   will   these   programs   work   accurately   for   small   number   of   instances   of  
these  relative  strangers?  And  what  about  the  success  of  these   programs  for  people  whose  images  change  
rapidly  such  as  infants  and  young  children?  

Another   place   where   machine   learning   has   been   applied   is   to   task   fragmentation.   TaskTracer  
(Dragunov   et   al.,   2005)   is   a   system   that   analyses   user   behaviours   in   an   attempt   to   organise   them  
according   to   activities.   One   major   problem   for   users   is   fragmentation,   whereby   resources   relating   to   a  
common   project   are   often   placed   in   separate   locations   by   applications.   Thus   the   emails,   spreadsheet,  
presentation   and   documents   for   a   project   may   all   be   in   different   folders,   making   it   hard   for   users   to  
collate  and  organise  task  related  materials  (Bergman  et  al.,  2004,  Boardman  and  Sasse,  2004).  Tasktracer  
addresses  this  by  analysing  temporal  access  patterns:  if  a  webpage,  document,  email  and  spreadsheet  are  
repeatedly   open   at   the   same   time,   then   the   system   infers   that   they   belong   to   the   same   task,   and  
 41  

constructs  a  virtual  folder  for  that  task.  The  user  can  choose  to  view  resources  in  the  virtual  folder  or  in  
their   regular   file   system,   but   the   benefits   of   the   virtual   folder   are   that   related   materials   are   clustered  
together.   Of   course   TaskTracer   suffers   from   the   same   problems   as   many   machine   learning   programs   in  
being  imperfect,  but  because  it  is  an  alternative  to  the  user’s  manual  files,  users  can  choose  to  use  it  if  
and  when  it  offers  benefits.  

Exploitation  

Technology   might   also   be   beneficial   for   various   aspects   of   exploitation.   One   obvious   area   is  
desktop   search.   Although   we   have   seen   that   desktop   search   is   currently   an   infrequent   way   to   access  
personal   data,   it   is   nevertheless   potentially   useful   as   a   ‘last   resort’   (Bergman   et   al.,   2008).   One   current  
problem   is   the   quality   of   desktop   search   which   generates   too   many   irrelevant   results.   Search   might   be  
improved   either   by   including   social   information   (e.g.   Millen   et   al.,   2007),   or   more   specific   data   about  
frequency  and  recency  of  document  access.      

There  might  also  be  different  ways  to  view  and  hence  access  our  personal  information  based  on  
automatically  captured  data.  One  approach  might  be  to  project  different  views  onto  the  user’s  data,  using  
readily   available   metadata   (time-­‐based,   social,   location).   These   views   are   not   meant   to   replace   existing  
folders   but   to   provide   alternative   ways   to   access   their   contents.   For   example,   we   have   seen   that   usage  
information  might  be  automatically  time  aligned,  so  that  all  resources  accessed  around  the  same  time  can  
be   accessed   together   (Dragunov   et   al.,   2005).   Radical   alternatives   such   as   Lifestreams   (Fertig   et   al.,  
1996a,b)  promise  to  replace  our  current  semantic  file  systems  with  operating  systems  that  are  purely  time  
based.   Other   radical   approaches   suggest   that   we   might   want   to   view   all   our   information   around   social  
relations   or   social   networks   (Nardi   et   al.,   2002,   Whittaker   et   al.,   2004),   and   these   systems   have   also  
proved   useful   as   alternative   email   clients.   And   other   hybrid   approaches   combine   search   with   temporal  
landmark   events   extracted   from   calendars   or   the   public   domain,   to   allow   people   to   access   documents  
using   these   events   as   landmarks   (Ringel   et   al.,   2003).   For   example,   a   user   might   be   able   to   look   at   the  
personal   information   that   they   accessed   shortly   before   a   business   trip   to   Boston   or   just   after  
Thanksgiving,   where   the   events   are   extracted   from   a   personal   calendar   (the   Boston   trip)   or   public  
resource  (Thanksgiving).    

Such  views  could  potentially  be  extended  to  other  types  of  metadata.  With  the  development  of  
cheap   sensors   it   is   now   possible   to   record   all   sorts   of   information   about   what   the   user   is   doing   at   any  
time.  Thus  it  might  be  possible  to  provide  information  about  where  the  user  was  when  s/he  produced  a  
document,   and   photos   or   other   recordings   may   be   available   about   other   activities   that   the   user   was  
engaged  in  when  that  document  was  being  worked  on  (Kalnikaite  et  al.,  2008,  2010).  For  example,  as  a  
user  might  recall  that  they  worked  on  a  presentation  for  a  business  trip  to  London,  and  a  locational  view  
might  allow  them  to  access  relevant  documents  by  this  cue.  Of  course  there  are  design  challenges  here:  
there  is  already  a  huge  amount  of  metadata  available  about  users’  activities,  and  interfaces  will  have  to  be  
carefully  designed  to  ensure  that  the  user  is  not  overwhelmed  by  this  richness.    

6.2  EMPIRICAL  AND  M ETHODOLOGICAL  ISSUES  

One  striking  observation  about  information  curation  is  that  we  know  very  little  about  it,  despite  
its   prevalence   in   everyday   computer   use   (Whittaker   et   al.,   2000).   Further,   most   previous   research   has  
focused   on   one   aspect   of   the   problem,   namely   management.   We   know   much   less   about   keeping   and  
exploitation   processes.   This   is   somewhat   ironic   given   the   vast   amount   of   research   effort   dedicated   to  
 42  

systems  and  tools  for  accessing  public  corpora.  More  critically  we  don’t  know  much  about  the  relationship  
between   different   aspects   of   information   curation,   and   perhaps   most   importantly   how   management  
strategies  influence  exploitation  success.  What  for  example  is  the  relationship  between  a  person’s  folder  
structure  and  their  ability  to  retrieve  and  access  files?  Much  more  research  is  needed  in  this  area.  We  also  
need   to   know   more   about   when   and   why   people   keep   or   delete   different   types   of  information,   exactly  
how  they  manage  and  reorganise,  as  well  as  the  different  methods  that  they  use  to  access  information.  At  
present  we  have  only  exploratory  studies  in  these  areas.    

There  are  several  practical  reasons  why  we  know  so  little.  The  first  is  that  it  is  extremely  hard  to  
gather   data   in   this   area.   To   better   understand   information   curation,   we   need   to   collect   data   about  
people’s  personal  information  habits.  This  is  potentially  intrusive,  as  it  might  require  logging  software  to  
be   installed   on   a   study   participant’s   machine,   or   manual   access   to   their   personal   data.   And   there   are   also  
problems   with   more   system   oriented   approaches:   if   we   want   to   study   the   efficacy   of   new   curation  
systems,  these  need  to  be  both  robust  and  fully  featured.  New  curation  software  needs  to  be  reliable  as  
people   use   it   on   a   regular   basis   for   everyday   work.   If   we   want   users   to   provide   feedback   about   a   new   file  
system,  email  client  or  web  bookmarking  system,  that  system  had  better  be  very  effective,  or  users  will  
quickly   switch   back   to   their   regular   software.   In   the   same   way,   the   new   system   had   better   offer   a  
comparable   set   of   features   to   users’   regular   software,   otherwise   participants   will   quickly   revert   to   that  
software  to  get  their  everyday  work  done  (Bellotti  et  al.,  2005,  Whittaker  et  al.,  2004).    

Further,  methods  for  evaluating  curation  systems  are  complex,  and  standard  techniques  cannot  
always   be   used   (Kelly,   2006,   Kelly   and   Teevan,   2007).   For   example   in   evaluating   information   retrieval  
systems   it   is   customary   to   use   standard   corpora   and   measures   such   as   precision   and   recall,   where  
documents   have   been   manually   tagged   for   relevance.   With   curation   systems   however,   we   need   to  
evaluate   systems   against   participants   own   information   as   the   use   of   public   data   would   be   meaningless.  
Further,  users  will  generate  their  own  access  tasks  exploiting  their  own  management  structures,  so  that  
methods  relying  on  relevance  metrics  generated  against  standard  corpora  cannot  be  applied.  In  part  this  
may   explain   why   promising   results   obtained   by   the   Machine   Learning   community   using   standard   public  
corpora  haven’t  yet  transferred  well  to  practical  curation  systems.  For  example,  new  algorithms  are  able  
categorise   email   data   in   standard   corpora   with   error   rates   around   10%.   Yet   we   don’t   know:   a)   what   error  
rates   users   will   tolerate   for   this   type   of   task   when   carrying   out   everyday   work;   and   b)   whether   similar  
performance  can  be  obtained  with  the  user’s  own  data.  In  our  own  work,  we  found  that  users  were  rather  
intolerant   of   automatic   methods   of   clustering   email   contacts,   instead   preferring   semi-­‐automated  
methods   to   organise   these   (Whittaker   et   al.,   2004).   More   studies   need   to   be   carried   out,   and   better  
evaluation   methods   developed   for   information   curation.   Elsewhere   we   have   advocated   that   the  
community   develop   a   set   of   reference   tasks   for   personal   information   management,   which   would   allow  
comparative  analysis  of  different  algorithms  across  a  common  set  of  user  tasks  (Whittaker  et  al.,  2000).          

7. SUMMARY    
This   review   has   argued   that   prevailing   views   of   our   information   behaviours   are   misleading.  
Instead   of   being   consumers   of   new   public   information,   people’s   informational   behaviours   are   closer   to  
curation,   in   which   they   keep   and   manage   personal   information   for   future   access.   We   have   outlined   a  
three  stage  model  of  the  curation  process,  reviewing  the  central  problems  of  keeping,  management  and  
exploitation   and   presented   relevant   data   for   each   stage   of   the   process,   concluding   with   an   overview   of  
outstanding  technical  and  empirical  questions.  In  general  users  tend  to  ‘overkeep’  information  with  the  
 43  

exception   of   contacts   and   web   pages.   When   organising   information   we   found   surprising   benefits   for   piles  
as   opposed   to   files,   although   organising   action-­‐oriented   information   remains   a   major   challenge.  
Exploitation   remains   reliant   on   manual   methods   such   as   navigation   despite   the   emergence   of   desktop  
search.  There  are  also  mismatches  between  people’s  organisational  structures  and  their  actual  retrieval  
requirements,  e.g.  for  email,  web  documents  and  photos.  There  are  a  number  of  new  technologies  that  
could  potentially  address  important  curation  problems,  but  implementing  these  in  user  acceptable  ways  
remains   a   challenge.   Finally   research   in   this   area   remains   in   its   infancy,   and   new   data   and   methods   are  
still  sorely  needed.    

REFERENCES  
Abrams,  D.,  R.  Baecker,  and  M.  Chignell.  1998.  Information  archiving  with  bookmarks:  personal  web  space  
construction  and  organization.  In  Proceedings  of  the  SIGCHI  Conference  on  Human  Factors  in  Computing  Systems,  
pp.  41–48.  New  York:  ACM  Press.  
Ackerman,  M.  S.  1998.  Augmenting  organizational  memory:  A  field  study  of  Answer  Garden.  ACM  Transactions  on  
Information  Systems  16(3):203–24.  
Ackerman,  M.  S.,  and  C.  A.  Halverson.  2004.  Organizational  memory  as  objects,  processes,  and  trajectories:  An  
examination  of  organizational  memory  in  use.  Journal  of  Computer  Supported  Cooperative  Work  13(2):155–90.  
Aula,  A.,  Jhaveri,  N.,  and  Käki,  M.  (2005)  Information  search  and  re-­‐access  strategies  of  experienced  web  users.  
Proceedings  of  WWW  2005,  May  10-­‐14,  2005,  583-­‐592.  
Baddeley,  A.D.  (1997).  Human  memory:  Theory  and  Practice,  Hove:  Psychology  Press.  
Balter,  O.  2000.  Keystroke  level  analysis  of  email  message  organization.  In  Proceedings  of  the  SIGCHI  Conference  
on  Human  Factors  in  Computing  Systems,  pp.  105–12.  New  York:  ACM  Press.  
Bälter,  O.,  and  C.  L.  Sidner.  2002.  Bifrost  inbox  organizer:  Giving  users  control  over  the  inbox.  In  Proceedings  of  
the  Second  Nordic  Conference  on  Human-­Computer  Interaction,  pp.  111–18.  New  York:  ACM  Press.  
Barreau,  D.  K.,  and  B.  Nardi.  1995.  Finding  and  reminding:  file  organization  from  the  desktop.  ACM  SIGCHI  
Bulletin  27(3):39–43.  
G.  Bell,  and  J.  Gemmell  2009.  Total  Recall:  How  the  E-­‐  Memory  Revolution  Will  Change  Everything,  Dutton.    
Bellotti,  V.,  N.  Ducheneaut,  M.  Howard,  and  I.  Smith.  2003.  Taking  email  to  task:  The  design  and  evaluation  of  a  
task  management  centered  email  tool.  In  Proceedings  of  the  SIGCHI  Conference  on  Human  Factors  in  Computing  
Systems,  pp.  345–52.  New  York:  ACM  Press.  
Bellotti,  V.,  N.  Ducheneaut,  M.  Howard,  I.  Smith,  and  R.  Grinter.  2005.  Quality  vs.  quantity:  Email-­‐centric  task-­‐
management  and  its  relationship  with  overload.  Human-­Computer  Interaction  20(1–2):89–138.  
Belkin,  N.  J.  (1980).  Anomalous  states  of  knowledge  as  a  basis  for  information  retrieval.  Canadian  Journal  of  
Information  Science,  5,  133-­‐143.  
Bentley,  F.,  Metcalf,  C.,  and  Harboe,  G.  (2006).  Personal  vs.  commercial  content:  the  similarities  between  
consumer  use  of  photos  and  music.  In  Proceedings  of  the  SIGCHI  Conference  on  Human  Factors  in  Computing  
Systems.  ACM,  New  York,  NY,  667-­‐676.    
Berlin,  L.  M.,  R.  Jeffries,  V.  L.  O’Day,  A.  Paepcke,  and  C.  Wharton.  1993.  Where  did  you  put  it?  Issues  in  the  design  
and  use  of  a  group  memory.  In  Proceedings  of  the  SIGCHI  Conference  on  Human  Factors  in  Computing  Systems,  pp.  
23–30.  New  York:  ACM  Press.  
Bergman,  O.,  Beyth-­‐Marom,  R.,  Nachmias,  R.  (2003).  The  user-­‐subjective  approach  to  personal  information  
management  systems,  Journal  of  the  American  Society  for  Information  Science  and  Technology,  v.54  n.9,  p.872-­‐
878,    
Bergman,  O.,  Beyth-­‐Marom,  R.,  Nachmias,  R.,  Gradovitch,  N.,  &  Whittaker,  S.  (2008).  Advanced  search  engines  
and  navigation  preference  in  personal  information  management.  Special  Issue  of  ACM  TOIS  on  Keeping,  Re-­‐
finding  and  Sharing  Personal  Information  26(4):  pp.  1-­‐24.  
Bergman,  O.,  Tucker,  S.,  Beyth-­‐Marom,  R.,  Cutrell,  E.,  and  Whittaker,  S.  2009.  It's  not  that  important:  demoting  
personal  information  of  low  subjective  importance  using  GrayArea.  In  Proceedings  of  the  27th  international  
Conference  on  Human  Factors  in  Computing  Systems  (Boston,  MA,  USA,  April  04  -­‐  09,  2009).  CHI  '09.  ACM,  New  
York,  NY,  269-­‐278.  
 44  

Blanc-­‐Brude,  T.  and  Scapin,  D.  L.  2007.  What  do  people  recall  about  their  documents?:  implications  for  desktop  
search  tools.  In  Proceedings  of  the  12th  international  Conference  on  intelligent  User  interfaces  (Honolulu,  
Hawaii,  USA,  January  28  -­‐  31,  2007).  IUI  '07.  ACM,  New  York,  NY,  102-­‐11.  
Boardman,  R.,  and  M.  A.  Sasse.  2004.  “Stuff  goes  into  the  computer  and  doesn’t  come  out”:  A  cross-­‐tool  study  of  
personal  information  management.  In  Proceedings  of  the  SIGCHI  Conference  on  Human  Factors  in  Computing  
Systems,  pp.  583–90.  New  York:  ACM  Press.  
Brewer  W  (1988)  Memory  for  randomly  sampled  autobiographical  events,  In:  U.  Neisser  &  E.  Winograd  (Eds.),  
Remembering  Reconsidered.  New  York:  Cambridge  University  Press,  pp  21-­‐90.  
Bruce,  H.,  W.  Jones,  and  S.  Dumais.  2004.  Information  behavior  that  keeps  found  things  found.  Information  
Research  10(1).  Available  at  http://informationr.net/ir/10-­‐1/paper207.html  
Capra,  R.,  and  M.  A.  Pérez-­‐Quiñones.  2005b.  Using  Web  search  engines  to  find  and  refind  information.  IEEE  
Computer  38(10):36–42.  
Catledge,  L.,  Pitkow,  J.,  1995.  Characterizing  Browsing  Strategies  in  the  World-­‐Wide  Web.  Computer  Networks  
and  ISDN  Systems  27(6):  1065-­‐1073.  
Civan,  A.,  W.  Jones,  et  al.,  2008.  Better  to  Organize  Personal  Information  by  Folders  Or  by  Tags?:  The  Devil  Is  in  
the  Details.  68th  Annual  Meeting  of  the  American  Society  for  Information  Science  and  Technology  (ASIST  2008),  
Columbus,  OH.  
Cockburn,  A.  and  S.  Greenberg.  Issues  of  Page  Representation  and  Organisation  in  Web  Browser-­‐Revisitation  
Tools.  Australian  J.  of  Info.  Systems,  7(2):120-­‐-­‐127,  2000.    
Cohen,  W.  1996.  Learning  rules  that  classify  email.  In  AAAI  Symposium  on  Machine  Learning  in  Information  
Access,  pp.  18–25.  Menlo  Park,  CA:  AAAI  Press.  
Crawford,  E.,  J.  Kay,  and  E.  McCreath.  2002.  An  intelligent  interface  for  sorting  electronic  mail.  In  Proceedings  of  
the  7th  International  Conference  on  Intelligent  User  Interfaces,  pp.  182–83.  New  York:  ACM  Press.  
Craik,  F.  I.  M.,  &  Lockhart,  R.  S.  (1972).  Levels  of  processing:  A  framework  for  memory  research.  Journal  of  Verbal  
Learning  and  Verbal  Behavior,  11,  671-­‐684.  
Cutrell,  E.,  S.  Dumais,  and  J.  Teevan.  2006a.  Searching  to  eliminate  personal  information  management.  
Communications  of  the  ACM  49(1):58–64.  
Cutrell,  E.,  D.  Robbins,  S.  Dumais,  and  R.  Sarin.  2006b.  Fast,  flexible  filtering  with  Phlat.  In  Proceedings  of  the  
SIGCHI  Conference  on  Human  Factors  in  Computing  Systems,  ed.  R.  Grinter,  T.  Rodden,  P.  Aoki,  E.  Cutrell,  R.  
Jeffries,  and  G.  Olson,  pp.  261–70.  New  York:  ACM  Press.  
Dabbish,  L.  A.,  R.  E.  Kraut,  S.  Fussell,  and  S.  Kiesler.  2005.  Understanding  email  use:  Predicting  action  on  a  
message.  In  Proceedings  of  the  SIGCHI  Conference  on  Human  Factors  in  Computing  Systems,  pp.  691–700.  New  
York:  ACM  Press.  
Dragunov,  A.N.,  Dietterich,  T.G.,  Johnsrude,  K.,  McLaughlin,  M.,  Li,  L.,  Herlocker,  J.L..  TaskTracer:  A  Desktop  
Environment  to  Support  Multi-­‐tasking  Knowledge  Workers.  International  Conference  on  Intelligent  User  
Interfaces.  p.  75-­‐82,  2005.  
Drew,  P.  R.  and  M.  D.  Dewe.  Special  collection  management.  Library  Management,  1992,  13(6):  8-­‐14.    
Donath,  J.  2004.  Visualizing  email  archives  (Draft).  Available  from  
http://smg.media.mit.edu/papers/Donath/EmailArchives.draft.pdf  
Dumais,  S.,  E.  Cutrell,  J.  Cadiz,  G.  Jancke,  R.  Sarin,  and  D.  Robbins.  2003.  Stuff  I’ve  Seen:  A  system  for  personal  
information  retrieval  and  re-­‐use.  In  Proceedings  of  the  26th  Annual  International  ACM  SIGIR  Conference  on  
Research  and  Development  in  Information  Retrieval,  pp.  72–79.  New  York:  ACM  Press.  
Ellis,  D.  and  M.  Haugan,  1997.  Modelling  the  information  seeking  patterns  of  engineers  and  research  scientists  in  
an  industrial  environment.  Journal  of  Documentation,  53(4),  384-­‐403.  
Elsweiler,  D.,  Baillie,  M.,  Ruthven,  I.  2008  Exploring  memory  in  email  refinding.  ACM  Trans.  Inf.  Syst.  26(4):  
(2008)  
Farina,  P.  A.  2005.  A  comparison  of  two  desktop  search  engines:  Google  Desktop  Search  (beta)  vs.  Windows  XP  
Search  Companion.  In  Proceedings  of  the  21st  Computer  Science  Seminar.  Hartford  CT.    
Fertig,  S.,  E.  Freeman,  and  D.  Gelernter.  1996a.  Finding  and  reminding  reconsidered.  SIGCHI  Bulletin  28(1):66–
69.  
Fertig,  S.,  E.  Freeman,  and  D.  Gelernter.  1996b.  Lifestreams:  An  alternative  to  the  desktop  metaphor.  In  
Conference  Companion  on  Human  Factors  in  Computing  Systems:  Common  Ground,  ed.  M.  J.  Tauber,  pp.  410–11.  
New  York:  ACM  Press.  
Fisher,  D.,  Brush,  A.  J.,  Gleave  E.,  and  Smith,  M.  (2006).  Revisiting  Whittaker  &  Sidner's  "Email  Overload";  Ten  
Years  Later,  CSCW  2006.  
 45  

Frohlich  D,  Kuchinsky  A,  Pering  C,  Don  A,  Ariss  S  (2002)  Requirements  for  photoware,  In:  Proceedings  of  the  
2002  ACM  Conference  on  Computer  Supported  Cooperative  Work  (CSCW’02),  New  Orleans,  Louisiana,  USA,  New  
York:  ACM  Press,  pp  166-­‐175.  
Gilbert,  D.  (2006).  Stumbling  on  Happiness,  Knopf.  
Golder,  S.,  Huberman,  B.  The  Structure  of  Collaborative  Tagging  Systems,  Journal  of  Information  Science,  
32(2):198-­‐208,  2006.    
Gonçalves,  D.,  &  Jorge,  J.A.  (2003).  In  An  Empirical  Study  of  Personal  Document  Spaces.  Paper  presented  at  the  
Proceedings  DSV-­‐IS'03,  Funchal,  Portugal.  
Gonçalves,  D.  and  Jorge,  J.  A.  2004.  Describing  documents:  what  can  users  tell  us?.  In  Proceedings  of  the  9th  
international  Conference  on  intelligent  User  interfaces  (Funchal,  Madeira,  Portugal,  January  13  -­‐  16,  2004).  IUI  
'04.  ACM,  New  York,  NY,  247-­‐249.  
Gwizdka,  J.  2004a.  Email  task  management  styles:  The  cleaners  and  the  keepers.  In  Proceedings  of  the  SIGCHI  
Conference  on  Human  Factors  in  Computing  Systems,  Exmanageed  Abstracts,  pp.  1235–38.  New  York:  ACM  Press.  
Gwizdka,  J.  2004b.  Cognitive  abilities  and  email  interaction:  Impacts  of  interface  and  task.  Doctoral  dissertation,  
University  of  Toronto,  Toronto.  
Henderson,  S.,  &  Srinivasan,  A.  (2009).  An  Empirical  Analysis  of  Personal  Digital  Document  Structures,  HCI  
International  2009.  San  Diego,  CA,  USA.  
Google.  2009.  Google  Desktop.  Retrieved  March  16,  2009,  from  http://desktop.google.com/  
Hearst,  M.  A.  1999.  User  interfaces  and  visualization.  In  Modern  information  retrieval,  ed.  R.  Baeza-­‐Yates  and  B.  
Ribeiro-­‐Neto.  Boston,  MA:  Addison-­‐Wesley.  
Jones,  W.  2004.  Finders,  keepers?  The  present  and  future  perfect  in  support  of  personal  information  
management.  First  Monday  9(3).  Available  at  http://www.firstmonday.dk/issues/issue9_3/jones/index.html  
Jones,  W.  2007.  Personal  information  management.  Annual  Review  of  Information  Science  and  Technology  
(ARIST)  41.  
Jones,  W.  2007.  Keeping  found  things  found:  The  study  and  practice  of  personal  information  management.  San  
Francisco,  CA:  Morgan  Kaufmann.  
Jones,  W.,  H.  Bruce,  and  S.  Dumais.  2003.  How  do  people  get  back  to  information  on  the  Web?  How  can  they  do  it  
better?  Paper  presented  at  the  9th  IFIP  TC13  International  Conference  on  Human-­‐Computer  Interaction  
(INTERACT  2003),  Zurich,  Switzerland,  September.  
Jones,  W.,  and  S.  Dumais.  1986.  The  spatial  metaphor  for  user  interfaces:  Experimental  tests  of  reference  by  
location  versus  name.  ACM  Transactions  on  Office  Information  Systems  4(1):42–63.  
Jones,  W.,  A.  J.  Phuwanartnurak,  R.  Gill,  and  H.  Bruce.  2005.  Don’t  take  my  folders  away!  Organizing  personal  
information  to  get  things  done.  In  Proceedings  of  the  SIGCHI  Conference  on  Human  Factors  in  Computing  Systems,  
pp.  1505–08.  New  York:  ACM  Press.  
Jones,  W.  and  Teevan,  J.  (2007).  Personal  Information  Management,  U  Washington  Press.  
Kahneman,  D.  and  Tversky,  A.  Prospect  theory:  An  analysis  of  decision  making  under  risk.  Eugene,  Ore.,  1979.  
Kalnikaitė,  V.,  Sellen,  A.,  Whittaker,  S.,  &  Kirk,  D.  (2010).  Now  Let  Me  See  Where  I  Was:  Understanding  How  
Lifelogs  Mediate  Memory.  To  Appear  in  Proceedings  of  CHI  2010,  ACM  Press,  New  York.    
Kalnikaitė,  V.  &  S.  Whittaker  (2008a).  Social  Summarization:  Does  Social  Feedback  Improve  Access  to  Speech  
Data?  In  Computer  Supported  Co-­‐operative  Work,  ACM  Press,  New  York.  
Kalnikaitė,  V.  &  Whittaker,  S.  (2008b).  Cueing  Digital  Memory:  How  and  Why  Do  Digital  Notes  Help  Us  
Remember?  In  Proceedings  of  Human  Computer  Interaction  (British  HCI  Conference),.    
Kalnikaite,  V.,  and  Whittaker,  S.  (2007).  Software  or  Wetware?  Discovering  When  and  Why  People  Use  Digital  
Prosthetic  Memory.  In  Proceedings  of  CHI07  Conference  on  Human  Factors  in  Computing  Systems,  71-­‐80,  New  
York:  ACM  Press.    
Kelly,  D.  2006.  Evaluating  personal  information  management  behaviors  and  tools.  Communications  of  the  ACM  
49(1):84–86.  
Kelly,  D.  &  Teevan,  J.  (2007).  Understanding  what  works:  Evaluating  personal  information  management  tools.  In  
W.  Jones  &  J.  Teevan  (Eds.),  Personal  Information  Management.  Seattle:  University  of  Washington  Press.  
Kidd,  A.  1994.  The  marks  are  on  the  knowledge  worker.  In  Proceedings  of  the  SIGCHI  Conference  on  Human  
Factors  in  Computing  Systems:  Celebrating  Interdependence,  ed.  B.  Adelson,  S.  Dumais,  and  J.  Olson,  pp.  186–91.  
New  York:  ACM  Press.  
Kobayashi,  M.  and  Takeda,  K.,  2000.  Information  retrieval  on  the  web.  ACM  Computing  Surveys  (ACM  Press)  32  
(2):  144–173  
Osburn,  Charles  B.,  and  Ross  Atkinson.  1991.  Collection  Management:  A  New  Treatise.  
Greenwich,  Connecticut:  JAI  Press.    
 46  

Kirk,  D.,  Sellen,  A.,  Rother,  C.,  and  Wood,  K.  (2006).  Understanding  “photowork”.  Proceedings  of  CHI  2006,  New  
York:  ACM  Press.  
Klimt,  B.,  and  Y.  Yang.  2004.  Introducing  the  Enron  corpus.  Paper  presented  at  the  First  Conference  on  Email  and  
Anti-­‐Spam  (CEAS  2004),  Mountain  View,  CA,  July.  Available  at  http://www.ceas.cc/papers-­‐2004/168.pdf  
Kobayashi,  M.  and  Takeda,  K.  2000.  Information  retrieval  on  the  web.  ACM  Computing  Surveys  (ACM  Press)  32  
(2):  144–173.  
Kuhlthau,  C.C.  1991.  Inside  the  search  process:  information  seeking  from  the  user's  perspective.  Journal  of  
the  American  Society  for  Information  Science,  42(5):  361-­‐371.  
Lansdale,  M.  1988a.  The  psychology  of  personal  information  management.  Applied  Ergonomics  19(1):55–66.  
Lansdale,  M.  1991.  Remembering  about  documents:  Memory  for  appearance,  format,  and  location.  Ergonomics  
34(8):1161–78.  
Lansdale,  M.,  and  E.  Edmonds.  1992.  Using  memory  for  events  in  the  design  of  personal  filing  systems.  
International  Journal  of  Man-­Machine  Studies  36:97–126.  
Lowe,  M.  2006.  Evaluation  of  desktop  search  applications.  Tech.  rep.  Kalio,  Sydney,  Australia.    
Lifestreams:  A  storage  model  for  personal  data.  1996.  ACM  SIGMOD  Bulletin,  March.  
Lifestreams:  Organizing  your  electronic  life.  1995.  AAAI  Fall  Symposium:  AI  Applications  in  knowledge  
navigation  and  retrieval,  November,  Cambridge,  MA.  
Mackay,  W.  E.  1988.  More  than  just  a  communication  system:  Diversity  in  the  use  of  electronic  mail.  In  
Proceedings  of  the  1988  ACM  Conference  on  Computer-­Supported  Cooperative  Work,  pp.  344–53.  New  York:  ACM  
Press.  
Malone,  T.  W.  1983.  How  do  people  organize  their  desks:  Implications  for  the  design  of  office  information  
systems.  ACM  Transactions  on  Office  Information  Systems  1(1):99–112.  
Marchionini,  G.  1995.  Information  seeking  in  electronic  environments.  Cambridge,  UK:  Cambridge  University  
Press.  
Marshall,  C.,  2008a.  Rethinking  Personal  Digital  Archiving,  Part  1:  Four  Challenges  from  the  Field,  in  DLib  
Magazine,  vol.  14,  no.  3/4,  Corporation  for  National  Research  Initiatives  (CNRI)/  D-­‐Lib  Magazine,  March  2008.    
Marshall,  C.,  2008b.  Rethinking  Personal  Digital  Archiving,  Part  2:  Implications  for  Services,  Applications,  and  
Institutions,  in  D-­‐Lib  Magazine,  vol.  14,  no.  3/4,  Corporation  for  National  Research  Initiatives  (CNRI)/  D-­‐Lib  
Magazine,  March  2008.  
Millen,  D.,  Yeng.,  M.,  Whittaker,  S.,  and  Feinberg,  J.  (2007).  Social  Bookmarking  and  Exploratory  Search.  In  
European  Conference  on  Computer  Supported  Co-­‐operative  Work,  179-­‐198.  Springer:  Amsterdam.  
Morris,  D.,  Ringel  Morris,  M.,  and  Venolia,  G.  2008.  SearchBar:  a  search-­‐centric  web  history  for  task  resumption  
and  information  re-­‐finding.  In  Proceeding  of  the  Twenty-­‐Sixth  Annual  SIGCHI  Conference  on  Human  Factors  in  
Computing  Systems  (Florence,  Italy,  April  05  -­‐  10,  2008).  CHI  '08.  ACM,  New  York,  NY,  1207-­‐1216  
Nardi,  B.,  S.  Whittaker,  E.  Isaacs,  M.  Creech,  J.  Johnson,  and  J.  Hainsworth.  2002.  ContactMap:  Integrating  
communication  and  information  through  visualizing  personal  social  networks.  Communications  of  the  
Association  for  Computing  Machinery  April,  pp.  89–95.  
Obendorf,  H.,  Weinreich,  H.,  Herder,  E.,  and  Mayer,  M.  2007.  Web  page  revisitation  revisited:  implications  of  a  
long-­‐term  click-­‐stream  study  of  browser  usage.  In  Proceedings  of  the  SIGCHI  Conference  on  Human  Factors  in  
Computing  Systems  (San  Jose,  California,  USA,  April  28  -­‐  May  03,  2007).  CHI  '07.  ACM,  New  York,  NY,  597-­‐606.  
Pazzani,  M.  J.  2000.  Representation  of  electronic  mail  filtering  profiles:  A  user  study.  In  Proceedings  of  the  5th  
International  Conference  on  Intelligent  Use  Interfaces,  pp.  202–06.  New  York:  ACM  Press.  
Petrelli,  D.,  Whittaker,  S.,  Brockmeier,  J.  (2008).  Autotopography:  What  can  Physical  Mementos  tell  us  about  
Digital  Memories?  In  Proceedings  of  CHI08  Conference  on  Human  Factors  in  Computing  Systems,  53-­‐62,  New  
York:  ACM  Press.    
Pirolli,  P.  (2007).  "Information  Foraging  Theory:  Adaptive  Interaction  with  Information."  
Pirolli,  P.,  &  Card,  S.  K.  (1995).  Information  foraging  in  information  access  environments.  In  Proceedings  of  the  
Conference  on  Human  Factors  in  Computing  Systems,  CHI  '95  (pp.  5158).  New  York:  Association  for  Computing  
Machinery.    
Pirolli,  P.,  &  Card,  S.  K.  (1999).  Information  Foraging,  Psychological  Review,  106,  643-­‐675.    
Ringel,  M.,  E.  Cutrell,  S.  T.  Dumais,  and  E.  Horvitz.  2003.  Milestones  in  time:  The  value  of  landmarks  in  retrieving  
information  from  personal  stores.  In  INTERACT’03,  ed.  G.  W.  M.  Rauterberg,  M.  Menozzi,  and  J.  Wesson,  pp.  184–
91.  Amsterdam:  IOS  Press.  
 47  

Robertson,  G.,  Czerwinski,  M.,  Larson,  K.,  Robbins,  D.  C.,  Thiel,  D.,  and  van  Dantzich,  M.  1998.  Data  mountain:  
using  spatial  memory  for  document  management.  In  Proceedings  of  the  11th  Annual  ACM  Symposium  on  User  
interface  Software  and  Technology  (San  Francisco,  California,  United  States,  November  01  -­‐  04,  1998).  UIST  '98.  
ACM,  New  York,  NY,  153-­‐162.  
Rodden,  K.,  and  K.  Wood.  2003.  How  do  people  manage  their  digital  photographs?  In  Proceedings  of  the  SIGCHI  
Conference  on  Human  Factors  in  Computing  Systems,  pp.  409–16.  New  York:  ACM  Press.  
Rosch,  E.  1978.  Principles  of  categorization.  In  Cognition  and  categorization,  ed.  E.  Rosch  and  B.  B.  Lloyd,  pp.  27–
48.  Hillsdale,  NJ:  Lawrence  Erlbaum.  
Rosch,  E.,  C.  B.  Mervis,  W.  Gray,  D.  Johnson,  and  P.  Boyes-­‐Braem.  1976.  Basic  objects  in  natural  categories.  
Cognitive  Psychology  8:382–439.  
Russell,   D.,   and   Lawrence,   S.   2007.   Search   everything.   In   Personal   information   management,   W.   Jones   and   J.  
Teevan,  eds.,  University  of  Washington  Press,  Seattle  and  London,  153-­‐166.    
Segal,  R.  B.,  and  J.  O.  Kephart.  1999.  MailCat:  An  intelligent  assistant  for  organizing  e-­‐mail.  In  Proceedings  of  the  
Third  Annual  Conference  on  Autonomous  Agents,  ed.  O.  Etzioni,  J.  P.  Müller,  and  J.  M.  Bradshaw,  pp.  276–82.  New  
York:  ACM  Press.  
Shannon,  C.,  and  Weaver,  W.,  1949.  A  mathematical  theory  of  communication.  University  of  Illinois  Press.  
Tang,  J.  C.,  Wilcox,  E.,  Cerruti,  J.  A.,  Badenes,  H.,  Nusser,  S.,  and  Schoudt,  J.  2008.  Tag-­‐it,  snag-­‐it,  or  bag-­‐it:  
combining  tags,  threads,  and  folders  in  e-­‐mail.  In  CHI  '08  Exmanageed  Abstracts  on  Human  Factors  in  
Computing  Systems  (Florence,  Italy,  April  05  -­‐  10,  2008).  CHI  '08.  ACM,  New  York,  NY,  2179-­‐2194.  
Tang,  J.  C.,  Lin,  J.,  Pierce,  J.,  Whittaker,  S.,  and  Drews,  C.  2007.  Recent  shortcuts:  using  recent  interactions  to  
support  shared  activities.  In  Proceedings  of  the  SIGCHI  Conference  on  Human  Factors  in  Computing  Systems  
(San  Jose,  California,  USA,  April  28  -­‐  May  03,  2007).  CHI  '07.  ACM,  New  York,  NY,  1263-­‐1272.  
Tauscher  ,  L.,  Greenberg,  S.  (1997).  How  people  revisit  web  pages:  empirical  findings  and  implications  for  the  
design  of  history  systems,  International  Journal  of  Human-­‐Computer  Studies,  v.47  n.1,  p.97-­‐137.  
Teevan,  J.,  C.  Alvarado,  M.  S.  Ackerman,  and  D.  R.  Karger.  2004.  The  perfect  search  engine  is  not  enough:  A  study  
of  orienteering  behavior  in  directed  search.  In  Proceedings  of  the  SIGCHI  Conference  on  Human  Factors  in  
Computing  Systems,  pp.  415–22.  New  York:  ACM  Press.  
Treisman,  A.  and  Gelade,  G.  A  feature-­‐integration  theory  of  attention.  Cognitive  Psychology  12  (1980),  97-­‐-­‐136.  
Venolia,  G.,  A.  Gupta,  J.  J.  Cadiz,  and  L.  Dabbish.  2001.  Supporting  email  workflow  (MSR-­TR-­2001-­88).  Redmond,  
WA:  Microsoft  Research.  
Venolia,  G.,  and  C.  Neustaedter.  2003.  Understanding  sequence  and  reply  relationships  within  email  
conversations:  A  mixed-­‐model  visualization.  In  Proceedings  of  the  SIGCHI  Conference  on  Human  Factors  in  
Computing  Systems,  pp.  361–68.  New  York:  ACM  Press.  
Wagenaar  W  (1986)  My  memory:  A  study  of  autobiographical  memory  after  six  years.  Cognitive  Psychology,  18,  
pp  225-­‐252.    
Wattenberg,  M.,  S.  Rohall,  D.  Gruen,  and  B.  Kerr.  2005.  Email  research:  Targeting  the  enterprise.  Human  
Computer  Interaction  20(1–2):139–62.  
Wen,  J.  2003.  Post-­‐valued  recall  Web  pages:  User  disorientation  hits  the  big  time.  IT  &  Society  1(3):184–194.  
Whittaker,  S.  2005.  Supporting  collaborative  task  management  in  email.  Human-­Computer  Interaction  20(1–
2):49–88.  
Whittaker,  S.,  V.  Bellotti,  and  J.  Gwizdka.  2006.  Email  in  personal  information  management.  Communications  of  
the  ACM  49(1):68–73.  
Whittaker,  S.,  Bellotti,  V.,  and  Gwizdka,  J.  2007.  Everything  Through  Email.  In  W.  Jones  and  J.  Teevan  (Eds.).  
Personal  Information  Management.  Seattle:  University  of  Washington  Press.  
Whittaker,  S.,  Bergman,  O.,  and  Clough,  P.,  2010.  Easy  on  That  Trigger  Dad:     A  Study  of  Long  Term  Family  Photo  
Retrieval.  Personal  and  Ubiquitous  Computing,  14(1),  31-­‐43.  
Whittaker,  S.,  and  J.  Hirschberg.  2001.  The  character,  value  and  management  of  personal  paper  archives.  ACM  
Transactions  on  Computer-­Human  Interaction  8(2):150–70.  
Whittaker,  S.,  Hirschberg,  J.,  Amento,  B.,  Stark,  L.,  Bacchiani,  M.,  Isenhour,  P.,  Stead,  L.,  Zamchick  G.,  &  Rosenberg,  
A.,  2002b  SCANMail:  a  voicemail  interface  that  makes  speech  browsable,  readable  and  searchable.  In  
Proceedings  of  CHI2002  Conference  on  Human  Computer  Interaction,  NY:  ACM  Press,  275-­‐282.    
Whittaker,  S.,  Jones,  Q.,  and  Terveen,  L.  2002a.  Contact  Management:  Identifying  Contacts  to  Support  Long  Term  
Communication.  In  Proceedings  of  Conference  on  Computer  Supported  Cooperative  Work,  216-­‐225.  New  York:  
ACM  Press.    
 48  

Whittaker,  S.,  Q.  Jones,  B.  Nardi,  M.  Creech,  L.  Terveen,  E.  Isaacs,  et  al.  2004.  Contactmap:  Organizing  
communication  in  a  social  desktop.  ACM  Transactions  on  Computer-­Human  Interaction  (TOCHI)  11(4):445–71.  
Whittaker,  S.,  and  C.  Sidner.  1996.  Email  overload:  Exploring  personal  information  management  of  email.  In  
Proceedings  of  the  SIGCHI  Conference  on  Human  Factors  in  Computing  Systems:  Common  Ground,  ed.  M.  J.  Tauber,  
pp.  276–83.  New  York:  ACM  Press.  
Whittaker,  S.,  L.  Terveen,  and  B.  A.  Nardi.  2000.  Let’s  stop  pushing  the  envelope  and  start  addressing  it:  A  
reference  task  agenda  for  HCI.  Human  Computer  Interaction  15:75–106.  
Wilhelm,  A.,  Takhteyev,  Y.,  Sarvas,  R.,  Van  House,  N.,  and  Davis,  M.  2004.  Photo  annotation  on  a  camera  phone.  In  
CHI  '04  Extended  Abstracts  on  Human  Factors  in  Computing  Systems.  CHI  '04.  ACM,  New  York,  NY,  1403-­‐1406.  
Wilson,  T.D.,  1981.  On  user  studies  and  information  needs.  Journal  of  Documentation,  37(1):  p.  3-­‐15.  
Wilson,  T.D.,  1994.  Information  needs  and  uses:  fifty  years  of  progress?  In  B.C.  Vickery,  (Ed.).  Fifty  years  of  
information  progress:  a  Journal  of  Documentation  review.  (p.  15-­‐51)  London:  Aslib.  
Wilson,  T.  1999.  Models  in  Information  Behaviour  Research.  Journal  of  Documentation.  55(3).  249-­‐270.  
49  

 
 50  

View publication stats