EdgeMaps: Visualizing Explicit and Implicit Relations

Marian Dörk, Sheelagh Carpendale, and Carey Williamson

University of Calgary, Canada

ABSTRACT
In this work, we introduce EdgeMaps as a new method for integrating the visualization of explicit and implicit
data relations. Explicit relations are specific connections between entities already present in a given dataset, while
implicit relations are derived from multidimensional data based on shared properties and similarity measures.
Many datasets include both types of relations, which are often difficult to represent together in information
visualizations. Node-link diagrams typically focus on explicit data connections, while not incorporating implicit
similarities between entities. Multi-dimensional scaling considers similarities between items, however, explicit
links between nodes are not displayed. In contrast, EdgeMaps visualize both implicit and explicit relations by
combining and complementing spatialization and graph drawing techniques. As a case study for this approach
we chose a dataset of philosophers, their interests, influences, and birthdates. By introducing the limitation of
activating only one node at a time, interesting visual patterns emerge that resemble the aesthetics of fireworks
and waves. We argue that the interactive exploration of these patterns may allow the viewer to grasp the
structure of a graph better than complex node-link visualizations.
Keywords: Information visualization, explicit and implicit relations, graph drawing, dimensionality reduction.

1. INTRODUCTION
An important goal of information visualization is to expose different types of relationships within abstract data.
Through interaction, the viewer can be enabled to find and understand connections between bits of information.
Relations can be explicitly present in a dataset as links that specifically connect information items or implicitly
by inferring relations based on similarity of attributes. For both types of relationships—explicit and implicit—
several visualization techniques have been proposed and refined over the recent years. Two of the most popular
techniques for visualizing relations are node-link diagrams (NLD) and multidimensional scaling (MDS). On the
one hand, NLD techniques are usually applied to explicit relations or connections that are visualized as edges
between nodes representing, for example, online communities, computer networks, or linked web pages. On the
other hand, MDS is typically used for implicit relations between documents or other types of multidimensional
data. MDS spatializes attribute similarities between items by placing similar items in close proximity to each
other and less similar items further apart.
While NLD and MDS techniques are widely used, they both have significant limitations with regard to
readability and interpretability of the resulting visualization. Layout algorithms for NLD visualizations are
typically optimized for reducing edge crossings, with the side effect that node positions are not utilized as a
meaningful visual variable. Furthermore, as the number of edges increases, it becomes hard to distinguish
directionality, if present, and identify high-degree nodes. The resulting visualization of an MDS algorithm, on
the other hand, lacks a guiding structure to put elements into context with each other, besides using position. It
is often difficult to understand the meaning of the positional proximity of elements. We argue that the limitations
of both techniques could be attributed to the fact that they are constrained to either explicit or implicit relations,
yet, many datasets feature both types of relationships.
In this paper, we explore how both explicit and implicit relations can be visualized as EdgeMaps ∗ , integrated
views that combine graph drawing and spatialization. EdgeMaps integrate NLD and MDS techniques utilizing
both visual linkage and proximity for the representation of complex—explicit and implicit—relations between
items. The intent behind this approach is to make effective use of visual variables that have been underutilized
in NLD and MDS techniques.
∗
Interactive demo available: http://mariandoerk.de/edgemaps/demo/
 ! TIME LINE ABOUT

SARTRE

00
00

00

00

20

40

50

90

10

30

40
00

20

60

80

00

20

50
00

00

00

00
10
13

16

17

17

17

18

18

19

19

19
18

18

18

18

19

19

19
-6

-5

-4

-2

KIERKEGAARD

BEAUVOIR

FOUCAULT
DEWEY
BUTLER
KITARO KANT
MERLEAU-PONTY
HERACLITUS
SPINOZA
HEGEL
HARTMANNBRADLEY
HABERMAS
MCTAGGART
MACHIAVELLI
BOOKCHIN
STIRNER
ENGELS CANTERBURY
MARX LOTZE
OAKESHOTT
KORSCH DIETZGEN
GRAMSCI
STEINERMCDOWELL
LAKATOSSCHELLING
DERRIDA DELEUZE DESCARTES
!I"EK POPPER
TAYLOR
DANTO
LUKáCS
CROCE
HEIDEGGER
BADIOU

LéVI-STRAUSS
ARISTOTLE
BRANDOM ADORNO

Georg Wilhelm Friedrich Hegel (IPA: [ˈgeɔʁk ˈvɪlhɛlm

LACAN ˈfʁiːdʁɪç ˈheːg#l]) (August 27, 1770 – November
LEGEND 14, 1831) was a German philosopher, and with Johann Gottlieb
Fichte and Friedrich Wilhelm Joseph Schelling, one of the
SIGNIFICANCE creators of German idealism.
degree of influence 1 Hegel influenced writers of widely varying positions,
54 including both his admirers (Bauer, Feuerbach, Marx,
Bradley, Dewey, Sartre, Küng, Kojève, !i"ek), and his
INFLUENCE A B C detractors (Schelling, Kierkegaard, Schopenhauer, Nietzsche,
A influenced B, and B influenced C Peirce, Russell). Hegel...

! INTEREST MAP ABOUT

00

00

00

20

30

60

20

40

70

90

00

20

30

40
30

00

00

50

60

80

10

50
00

00

00

00

0
10
12

14

16

17

17

17

18

18

18

18

19

19

19

19
16

17

18

18

18

18

19

19
-6

-5

-4

-3

BL NA
EMAR

SP

GO TZ

ST

SC
BU

ST

CA

RA

DE
FO UZ

W
D
GEUR IAM
BL ARI S
ROO TTDA

VA

CA

BU
M
PA

BE

BEER
SP CAL

HEITT
HE
P

SC

AD

SA N O
Lé RTR
W
NI
PL

KA

IL
A
EI

RA
IR

TT
H
BEEL

LE
UC E

PU

TL
VI E
S

NJ

AULE
PEMCL

RTM RI
RD SC
E
RA

HO

OR

NC

L
IN

ID GE

O
AT

NE

US

LS

IM

E
A

US
RER

TO
NT
DEONITU

VOAU

AU

Y
ON H
OZ

EG N

HO

R
M

A
PE

O
O

S
S
IN

LT
IR - P
CILDES

T
A

GEST
NH
ESS

AU

R EIN

ON
E
E

TY
R

Friedrich Wilhelm Nietzsche (October 15, 1844 – August

LEGEND 25, 1900) (German pronunciation: [ˈfʁiːdʁɪç
ˈvɪlh!lm ˈniːtʃ!]) was a nineteenth-century German
SIGNIFICANCE philosopher and classical philologist. He wrote critical
1 texts on religion, morality, contemporary culture,
degree of influence
54 philosophy, and science, using a distinctive German language
style and displaying a fondness for metaphor and aphorism.
INFLUENCE A B C Nietzsche's influence remains substantial within and beyond
A influenced B, and B influenced C philosophy, notably in existentialism and postmodernism....

Figure 1. Visualizing influence relations between philosophers using an interest map (top) and a timeline (bottom).

As a case study for this paper, we have chosen a dataset of philosophers from the Freebase† data community.
While there are many biographical records associated with philosophers, a particular interesting aspect of this
dataset is the existence of influence connections between philosophers, which are a type of explicit relationship.
On the other hand, birthdates, interests, and professions can be seen as attributes that indicate implicit relations
†
http://www.freebase.org
between philosophers. We chose these dimensions as they provide a compelling use case for the visualization
of explicit and implicit relationships. For example, visualizing influences between philosophers as edges may
indicate which philosophers had more impact, yet, it is not possible with these links alone to see the extent of
the impact a philosopher had across time and interests. However, positioning philosophers based on birthdates
or interests can make the extent of influence much clearer (see Figure 1).
The remainder of the paper is structured as follows. First, we will provide a brief overview of prior work on
visualizing complex, multidimensional data (Section 2). Then we formulate our main design goals (Section 3)
and describe the dataset we use to exemplify the techniques introduced by this work (Section 4). Based on
the design goals and the example dataset, we introduce EdgeMaps as a method for visualizing both explicit
and implicit relations (Section 5). We discuss novel ways for exploring complex data relations (Section 6) and
critically assess limitations and open questions of this work (Section 7), before we conclude this paper.

2. RELATED WORK
As visualizing relationships is at the heart of information visualization (infovis), our work builds upon many
previous contributions in the field of infovis, in particular with regard to the use of visual variables, graph
drawing methods, and casual visualization.
While not part of his visual information-seeking mantra (“Overview first, zoom and filter, then details-on-
demand”), Shneiderman notes the challenge of being able to explore relationships between information items.1
He stresses the importance of interaction for relating data entries, however, equally if not more important are
the appropriate visual representations of different types of relations. To think about representing relationships
visually it is worth considering the visual variables that are at our disposal. In Semiology of Graphics, Bertin
distinguishes between eight visual variables: size, value, texture, colour, orientation, shape, and the two dimen-
sions for the position on the plane.2 MDS renderings use planar position as the primary visual variable, while
NLDs typically rearrange position in order to minimize edge crossings. Stone makes the case that colour can
make visualizations more effective and beautiful when used well.3 She shows how colour can be used for labelling
and quantifying data. It would be interesting to explore the use of colour for conveying similarity between items
as a degree of association or disassociation in Bertin’s terms.
There has been extensive research on drawing and interacting with NLDs,4 often with the goal of minimizing
edge crossings, which is one of several geometrical and graph-theoretical metrics for graph aesthetics.5 Recent
additions to this research include EdgeLens, a technique for interactively exploring overlapping edges,6 and Edge
Bundles, a method for combining edges with similar paths.7 While these contributions significantly improve
the readability of large NLDs, we argue that we should de-emphasize the role of edges and acknowledge more
the contextual attributes of nodes. In an early work on computer network visualizations, visualization of edge
attributes (e. g., flow, capacity, utilization) was seen as more important than the actual network topology.8 As
part of a social network visualization it was shown how the visual representation of number of friends, gender,
and community structure enriches the NLD and allows for interactive filtering.9
While conventional NLD techniques focus almost entirely on explicit relations, MDS can be seen as a com-
plementary approach focussing on proximity as a visual representation of implicit relations or similarity. MDS
has been used for document visualizations with the goal to visually convey “thematic patterns and relationships”
of text collections.10 While the idea of spatializing document collections based on their similarities or differences
is promising, the resulting galaxies and themescapes still appear abstract and difficult to interpret. An approach
to making MDS more interactive focused on steering the algorithm, but did not look at using interactivity to
make the MDS view more meaningful and accessible.11
Besides linkage and similarity, another important type of relation is based on the temporal dimension and
references between temporally structured items. Considering that time is generally seen as a linear dimension, the
challenge is to visualize relations and references between items that are mapped onto a linear axis. Arc diagrams
show cross-references along a linear axis by adding visual semicircles to linear visualizations of documents, music
pieces, and DNA sequences.12 Further exploring the possibilities of such arcs, arcs have also been used to visualize
email threads.13 It was shown how the combination of arcs displayed above and below the main axis improved
readability.
 Several infovis contributions have looked at enriching and combining techniques for visualizing different types
of relations. For example, a visual document hierarchy was accompanied with arcs representing cross-references
between different sections.14 Furthermore, NLDs were made more readable by assigning nodes into multiple
regions and allowing for interactive edge filtering.15 Another approach suggests the three-dimensional juxtapo-
sition of different visualizations on panes and linking corresponding nodes using edges between the panes.16 All
these techniques underline that there is a need for integrating explicit and implicit relations and enabling their
interactive exploration, however, there is a tendency to emphasize one over another.

3. DESIGN GOALS
The motivation behind this work is the multitude of datasets that feature both explicit and implicit relations
and preliminary research on complementing the visualization of one type of relation with aspects of the other.
For this work, we understand explicit relations as data relations that specifically connect data entities and are
already present in the dataset. As implicit relations we see data relations that are not defined in the dataset
and need to be inferred based on similarities between data entries.
Explicit linkage between items has been the mainstay of graph drawing research. On the other hand, there
are numerous implicit similarities based on different kinds of parameters and dimensions that can be used for
data spatialization. With this work we are exploring the space of integrating the visualization of both explicit
and implicit data relations in order to reveal previously unseen patterns. In particular, we aim at supporting
people in viewing complex datasets and exploring relationships between information entries in a pleasing and
engaging way. This translates into the following design goals:

• Integrate multiple relationships. Explicit and implicit relations should be visually represented as
linkage and layout in ways that mutually support each other.
• Show invisible data patterns. By integrating explicit and implicit data connections, the visualization
should provide novel, interesting patterns that were not visible before.
• Support serendipitous exploration. The visualizations should allow viewers to find unexpected insights
and easily follow their interests. The interactivity necessary should be effortless without requiring training.
• Display additional information. The interface should provide detail-on-demand operations allowing
the viewer to learn more about particular data entries and go back to the data source.
• Provide aesthetic visuals. The colours, shapes, and transitions used by the visualization should satisfy
both utility and visual appeal, making the interaction pleasant and evoking curiosity.

4. DATASET AND DIMENSIONS

We parsed data on philosophers from the Freebase website using their HTTP-based API. Freebase offers struc-
tured information about many entities. In the case of philosophers, Freebase provides data about interests,
professions, birthdates, influence connections, and other relations from Wikipedia. For the purpose of this pa-
per, we have constrained the dataset to philosophers that influenced at least one other philosopher, resulting in
142 philosophers. For each philosopher we store the name, birthdate, description, an image, interests, and pro-
fessions. Furthermore, we store the directed influence links between philosophers, which we consider as explicit
relations. In contrast, we will use interests and professions for computing similarities between philosophers as
implicit relations. The birthdates can also be regarded as another type of implicit relation as it implicitly links
philosophers of similar epochs. For storing these records, we used a MySQL database that is easily accessible
from the server-side part of this visualization, which was written in PHP.
Using these attributes, several interesting dimensions can be inferred and explored. For example, we can
calculate the degree of influence as a measure of significance using the sum of outgoing influence. In other words,
the more philosophers a particular philosopher has inspired and influenced, the more significant this philosopher
is. Based on birthdates, philosophers can be grouped into similar epochs. Likewise, using the interests and
professions, philosophers can be grouped by their “trade”. Combining birthdates and interests/professions with
the influence connections, one can look at the impact of a philosopher across time and trade. So, for example, as
Nietzsche was considered influential by many subsequent philosophers, his impact is one of great temporal extent
(see Figure 1, bottom). Similarly, considering all the philosophers that were influenced by Hegel, his impact
extended over a wide range of interests and professions (see Figure 1, top). These relations, however, are not
easily accessible by just looking at tables or reference pages on websites such as Freebase or Wikipedia.

5. VISUALIZING EXPLICIT AND IMPLICIT RELATIONS

To develop a way of representing both explicit and implicit relations, we introduce EdgeMaps as a visualization
method that integrates spatialization and graph drawing techniques. EdgeMaps encode explicit relations as
curved edges and implicit relations as node position. Other visual variables are used to double-encode these data
relations and introduce additional information such as directionality and distinctness.

5.1 Implicit Relations as Layout

To represent the implicit relations, we designed two general layouts: an interest map and a timeline (see Figure 2).
While both visualization layouts represent philosophers as nodes and influences as links, as we will describe in
more detail later, the layouts differ in the way the positions on the plane are utilized. The interest map represents
the similarity of philosophers based on their philosophical interests and professional occupations. The timeline
uses birthdates as an ordering criteria to arrange philosophers along a temporal axis.

Interest Map Timeline

Figure 2. Layouts for philosopher visualizations based on common interests and professions (left) and on birthdates (right).

Interest map. The node positions for the interest map were computed using the MDS function of R‡ ,
a statistical software environment. Before that, interests and professions data had to be transformed into a
vector model for which the Euclidian distances could be used by the MDS algorithm. The result is a pair of
x,y coordinates for each philosopher, with each coordinate value between -1 and 1. Depending on the size of
the window, these coordinates are then scaled to the actual display resolution. However, the aspect ratio of
the MDS output is not modified based on the window, since the proximity of the resulting plane configuration
is generated on the basis of similarity. Stretching the layout would confound the representation of similarities,
which are mapped to positions on the plane using their Euclidian distances. As there may be philosophers
with identical attributes, the MDS algorithm could return items with the same positions, posing an occlusion
problem. Considering that MDS is an approximation after all, the current layout method positions overlapping
nodes slightly apart so that they are still close but not occluding each other.
Timeline. The timeline maps birth year to position on the plane. Initial trials with a simple time axis
failed, because there were dense clusters of philosophers in some epochs and very sparse or empty periods during
other times. For the 142 philosophers, we decided to neglect exact temporal distances and focussed on temporal
sequences of philosophers instead. This still allows for relative temporal comparisons of ‘earlier’ and ‘later’
philosophers and, at the same time, accommodate all philosophers along the axis. The result of this ‘stringing’
of nodes along the axis has the effect that certain periods take up much more display space than others. A time
legend displayed on top of the time view is designed to convey this temporal folding.
‡
http://www.r-project.org/
5.2 Explicit Relations as Curved Edges
The layouts for topical and temporal similarities—interest map and timeline—represent only the implicit relations
between philosophers. In order to represent influences between philosophers as explicit relationships, edges are
drawn between the nodes. If all influence edges were to be shown for all philosophers, there would be far too
many edges to be actually readable, let alone interpretable. However, by activating only one philosopher at a
time, it is possible to read individual edges and differentiate between two types of influence:

• Incoming influence. The philosopher is inspired by previous philosophers and builds upon their work.
• Outgoing influence. The philosopher has affected other philosophers who built upon his/her work.

A B C

Figure 3. With node B being selected, there are two types of edges: incoming (left) and outgoing (right).

To visually differentiate between these two types of edges, we used the following visual parameters: curvature,
directionality, shape, value, and colour (see Figure 3).
Curvature. Edges are curved depending on their distance between the connected nodes. In the interest map
this curvature is smaller than in the timeline view. Incoming edges are curved downward and outgoing edges
are curved upward. The idea is that incoming influence stands for the foundation upon which a philosopher
builds their work. For the outgoing influence, the edge is curved upward, as a visual depiction of outreach
beyond previous work. Taken together, both types of edges form a wave-like shape when nodes are arranged in
a sequential order.
Directionality. Incoming and outgoing edges differ also with regard to how directionality is represented.
As we assume that only one philosopher is activated at a time, it is evident that this philosopher’s node will
have multiple incoming and outgoing edges, while all other philosophers will have at most two associated edges.
This means to represent edge directionality, there is much less clutter around the unselected philosophers that
are associated with the active philosopher. Therefore the incoming edges have arrow-like cuttings at their source
nodes, whereas the outgoing edges coming from the active philosopher have arrows at the destination nodes.
This way the edge endings at active philosophers are simple and thin allowing for many discernible edges.
Shape. An additional way to visually differentiate the edges is by their shapes. Incoming edges are drawn
significantly thicker than the outgoing edges. While there is no inherent reason why incoming edges should be
thicker than outgoing edges, giving them these two distinct shapes allows for easy distinction with multiple edges
displayed at the same time.
Opacity. To balance out the different visual weights resulting from thick and thin edges, the opacity of the
incoming edges is decreased. Using opacity instead of adjusting brightness for the incoming edges also reduces
the occlusion of other elements due to incoming edges.
Colour. Since the nodes representing philosophers have different colours, the outgoing influence edges take
on the colour of the active philosopher, and the incoming edges share the colour with the originating node. The
mapping of position to colour is discussed in the following section.

5.3 Encoding Redundancy with Size and Colour

Besides showing influence connections for the active philosopher, the significance of a philosopher should be
immediately visible without any interaction. For this, we define significance as the relative degree of outgoing
influence, since a more significant philosopher arguably had more influence on their philosopher peers and suc-
cessors. To visually represent this measure, we encoded significance as the size of each philosopher’s node. While
we experimented with different types of glyphs, we decided that simple circles with varying sizes were the most
appropriate, with which additional clutter could be avoided.
 Furthermore, we decided to use colour to double-encode similarity based on philosophers’ interests and
professions. Since the output of the MDS algorithm provides a spatialization of this relationship, we used it
in combination with the HSV (hue, saturation, value) colour space as the basis for the colour calculation (see
Figure 4, left). We decided to map broad interest regions to hue and the distance to the centre to saturation,
while keeping the brightness constant. The idea behind this encoding is that the further out a philosopher is
located, the more distinct this philosopher is from all other philosophers. Translating this notion of distinctness
to a colour space, it would seem intuitive that the more distinct items would be more saturated and the more
common items would be less saturated. However, keeping the value constant means that either the nodes in the
centre would be too bright or the nodes on the periphery too dark. Therefore, we took the distance-to-centre
measure into account for the value resulting in well-visible, grey-like nodes in the centre and more colourful nodes
towards the outside.

! TIME LINE
SARTRE

00

00

20

50

60

00

30

40
00

20

40

00

80

90

10

20

50
00

00

00
00
90°

0
90
13

17

18

18

18

19

19

19
16

17

17

18

18

18

19

19

19
-6

-5

-4
-2
PETTY

BEAUVOIR

hue ARENDT
FOUCAULT
CAMUS KANT
saturation BARTHES MERLEAU-PONTY
SMITH
HEGEL
SPINOZA
HABERMAS
WEIL

BOOKCHIN
STIRNER
0° 180° COHEN
CASTORIADIS
BERLIN
MARX MCCABE
MOSCA RAWLS DIETZGEN BENJAMIN
BALIBAR
GRAMSCI
KORSCH
DERRIDA
!I"EK LAKATOS
ALTHUSSER LUKáCS
CROCE LYOTARD

BADIOU PUTNAM
LéVI-STRAUSS
RANCIèRE
ADORNO
Karl Heinrich Marx (May 5, 1818–March 14, 1883) was a
CHOMSKY German philosopher, political economist, historian,
sociologist, humanist, political theorist, and revolutionary
LEGEND
credited as the founder of communism.
Marx summarized his approach to history and politics in the
SIGNIFICANCE
1 opening line of the first chapter of The Communist Manifesto
1 degree of influence
54 (1848): “The history of all hitherto existing society is
47 270° the history of class struggles.” Marx argued that
INFLUENCE A B C capitalism, like previous socioeconomic systems, will
B C A influenced B, and B influenced C produce...
was influenced by C

Figure 4. Colours are derived from MDS positions using distance from the centre for saturation and the angle around the
centre for the hue (left). Incoming edges are coloured according to the source node and outgoing edges have the colour of
the active node (right).

Based on this approach, every node receives a distinct colour, which is also used for the outgoing influence
edges associated with this node. As shown in Figure 4 right, the active philosopher’s outgoing edges have the
colour of this philosopher and the incoming edges have the respective colours of their philosophers. This further
helps to distinguish the edge types and associate edges with their corresponding source nodes.

5.4 Yarn Balls vs. Fireworks and Waves

After having discussed the representation of implicit and explicit relations individually, we will discuss in the
following how these representations come together in novel formations.
While exploring several design options, we examined the possibility of activating multiple philosophers at the
same time. The result is an example of the “yarn ball” effect for complex NLDs, conveying neither overview nor
structure (see Figure 5, left). In contrast, displaying only the edges associated with one individual node opens up
novel possibilities (see Figure 5, middle and right) for graph visualization and interaction. As discussed before,
this way it becomes possible to actually distinguish between edge types and directions. Furthermore, considering
that the node size reflects the number of outgoing edges, one could argue that differing node sizes may be a more
effective way to convey general overviews over a dataset. Instead of displaying many edges that make it even
harder with increasing number of nodes and edges, it seems to be more effective to use colour, position, and size
for nodes to provide context and overview (see Figure 4, left and Figure 6, top left).
An interesting structure emerges when selecting a philosopher in the interest map with incoming and outgoing
edges rendered in patterns of fireworks (see Figure 5). Less significant philosophers (e.g., Lacan) result in more
 ! TIME LINE
! TIME LINE

SARTRE
SARTRE

STRAUSS
KIERKEGAARD STRAUSS
KIERKEGAARD

AQUINAS AQUINAS
PETTY

SCOTUS

BEAUVOIR WEININGER BEAUVOIR WEININGER

HUME HUME
ARENDT ARENDT
FOUCAULT
FOUCAULT
MILL MAIMON
MAIMON
CAMUS
MERLEAU-PONTY KANT MERLEAU-PONTY KANT
BRADLEY HEGEL HEGEL BRADLEY HEGEL
HERACLITUS
SMITH SPINOZA
BARTHES WEIL
MISES SPINOZA BARTHES
MISES
HAREREINHOLD
CAPUTO SOCRATES
OCKHAM GORDON
REINHOLD
GORDON ENGELS
STIRNER
ENGELS HARE SCHELER CANTERBURY
HOBBES SCHOPENHAUER
SCHOPENHAUER
MARX LéVINAS
BERLIN
FOOT
HAYEK MALEBRANCHE MARX HAYEK MALEBRANCHE
SPIR
SPIR RAWLS RENOUVIER
RICOEUR
RAWLS RENOUVIER
RICOEUR
BLANCHOT BENJAMIN LOCKE
BALIBAR
GUATTARI KORSCH
LOCKE GUATTARI SELLARS
AYER
GRAMSCI BERGSON
AYERBERGSON
PARMENIDES SELLARS
DAVIDSON DAVIDSON
SCHELLING
CARNAP
SCHELLING DERRIDA
DERRIDA CARNAP DELEUZE
POPPER DESCARTES
POPPER
DELEUZE
LAKATOS DESCARTES
PLATO
BUBER BUBER BAUMGARTEN
! INTEREST MAP ALTHUSSER
BAUMGARTEN
CROCE BACHELARD ALTHUSSER LUKáCS
CROCE
HUSSERL
LYOTARD STRAWSON
BERKELEY
LUKáCS HUSSERL RORTY
LYOTARD BERKELEY
STRAWSON
EMPEDOCLES RYLE NIETZSCHE HEIDEGGER
NIETZSCHE HEIDEGGER HEIDEGGER
BADIOU BADIOU PUTNAM
DEMOCRITUS
ANAXIMANDER PUTNAM WITTGENSTEIN
WITTGENSTEIN
LéVI-STRAUSS
LéVI-STRAUSS PASCAL
VATTIMO ARISTOTLE
RANCIèRE RANCIèRE WHITEHEAD
ADORNO
ADORNO ADORNO
WHITEHEAD
00

00

30

30

70

00

20

40

50

60

80

90

00

20

30

40
00

00

00

1
MACH
20

13

16

16

17

17

17

17

18

18

18

18

18

18

18

19

19

19

19

19
-6

-5

-4

CHOMSKY
MACH SCHILLER
CHOMSKY
SCHILLER

LEGEND LACAN
LEGEND Jacques-Marie-Émile Lacan (French pronounced [!ak GADAMER
Imma
Aristotle
Karl Heinrich
(Greek:
Marx
()*+,-,./01,
(May 5, 1818–March
Aristotél2s)14, 1883) was a lak!"]) (April 13, 1901 – September 9, 1981) was a
GADAMER 12 Fe
SIGNIFICANCE
Figure 5. Displaying influence edges in the
SIGNIFICANCE
1
interest Friedrich
René
Immanuel
(384
map
25,
(31
12
German
Plato
Descartes
BCWilhelm
–philosopher,
sociologist,
1900)
March
February
Kant
322 BC)
for
and(German
1596
teacher
(IPA:
(French
Nietzsche
humanist,
1804)
– 11
was
[!'manu"l
five political
pronunciation:
a Greek
pronunciation:
of
wasFebruary
Alexander
(October
political
economist,
kant];
an 18th-century
1650),
philosopher,
philosophers
theorist,
the
15,
22[$&ne
April
[#f$i%d$!ç
Great.
also
1844
historian,
and
German
known
deka$t]),
1724
–
a August
student
–
Herevolutionary
wrote
philosopher
as on
of
many (left) and one philosopher at a time (middle
Renatus
degree of influence
French psychoanalyst and psychiatrist who made prominent
1
contributions to psychoanalysis, philosophy, and literary
from t
Russi
#v!lh&lm
Cartesius
from
subjects,
credited
the#ni%t'&])
Prussian
(Latinized
as
including
thewas founder
city
physics,
aform),
nineteenth-century
of of
Königsberg
was
communism.
metaphysics,
a French (now philosopher,
German
poetry,
Kaliningrad,
theater, 47
degree of influence theory. He gave yearly seminars, in Paris, from 1953 until thinke
and right). With a single activated philosopher, influence
47 philosopher
mathematician,
Russia).
music,
Marx summarized
texts
adult
thinkers
biology
opening
on
logic,
Heand
lifereligion,
and
in
ofline
the
modern
edges
isrhetoric,
regarded
scientist,
classical
zoology.
ofDutch
hispolitics,
morality,
theEurope
first
approach
and
as
Republic.
chapter
arguably
philologist.
one
writer
government,
oftothe
contemporary
andHe
who
history
ofofthe
Hemost
has
wrote
spent
Thelate
been
and
ethics,
influential
culture,
Communist
critical
most ofresemble
politics
Enlightenment.
his
dubbedManifesto
the
in the the shape of fireworks.
1981, Less
most influencing France's influential
intellectuals in the 1960s Kant c
philosophy,
"Father
Kant
Together
(1848): and the 1970s, especially the post-structuralist which
INFLUENCE B created
of“The
Modern
with
andahistory
Plato
science,
newPhilosophy,"
widespread
and
of all
using
Socrates
hitherto
aand
distinctive
perspective
(Plato's
existing
much ofteacher),
German
society
subsequent
in philosophy
islanguage A C
philosophers
INFLUENCE B haveAsparser edge patterns (middle) than
style
Western
which
Aristotle
the and
C more
history
influenced
philosophy
displaying
is one
of class
of influential
philosophy
the
ais
struggles.”
fondness
a
most
LEIBNIZresponse
important
through
for
Marxto
metaphor
his
the
argued
founding
writings,
21st and
that ones
Century.
figures
which
aphorism.in
He (right). philosophers. His interdisciplinary work is Freudian,
featuring the unconscious, the castration complex, the
B influenced A, and B was influenced by published
C to
LEIBNIZ
also p
works
Nietzsche's
continue
also
Western
capitalism,
philosophy.
be
influence
like
studied
important
previous
remains
He
closely
was
works
socioeconomic
substantial
the
to of
this
first
epistemology,
day.
to create
systems,
within
In particular,
aand
aswill
beyond
well as
B influenced A, and B was influenced by C philosophy,
his
works
comprehensive
produce...
Meditations
relevant
notably
on
to
system
religion,
First
in existentialism
Philosophy
of law,
Western
andcontinues...
and
philosophy,...
history....
postmodernism.... ego;...

! INTEREST MAP

G ER
D

RA
AD

BA

AL

BA
LA

Lé
HE

UA R
HE

NC
TH

DIO
RT
O

VI
CA

TTIDA
RN
ID

Iè
US
HE
G

-S

U
AR

RE
N

O
EG

SE
TR
EL

I
R
AU
G
ER
00

00

30

00

20

30

70

00

20

40

50

60

80

90

00

10

20

30

40
SS
00

00

00

0
20

13

16

16

17

17

17

17

18

18

18

18

18

18

18

19

19

19

19

19
-6

-5

-4

LEGEND Jacques-Marie-Émile Lacan (French pronounced [!ak

lak!"]) (April 13, 1901 – September 9, 1981) was a
French psychoanalyst and psychiatrist who made prominent
SIGNIFICANCE
1 contributions to psychoanalysis, philosophy, and literary
degree of influence
47
theory. He gave yearly seminars, in Paris, from 1953 until
1981, most influencing France's intellectuals in the 1960s
and the 1970s, especially the post-structuralist
philosophers. His interdisciplinary work is Freudian,
INFLUENCE A B C
featuring the unconscious, the castration complex, the
B influenced A, and B was influenced by C ego;...
BAHU

ST
REAIM

RE

SP

BR

CR

W B
M EINER

DA

CH U AM
RI LLA

PU UZ
M

SE
SCIN O

SHTA

HAR
BE

LU

PO

M
M C ZA

BU

AD PER

AR

BE RL

AY

RA

LY

DE RD

FOTN
SC G

CA E
E

BE SS
HU
AQ

LO NO

L EB

SA O
DE

HE TG
SC ELL L

W
SP

AD
HE

KI

NI

IS IN
KA

M
NG
AL K

AC

CO RS
UM M
EI RA

IR

H
NOE

AE

VID
RYAE
O

O CA
HIL

RAEW
RD

O
AU EA

ER R ON
Ká

EN

LE
ES G
RK

P
H E

RN GG TE
DL

AM
RG E

IT

CE

M U
I

RT
HON

TA
AR
HO ING
BN NC
UI

RN
H
EL

T
SC

ID EN
G E
I

LS
UKS

EU
UV

SO

SK L
LELD

VO U -
C

DT
EH
EY
R

SO
EL

AR

TZ

ER
AP
N
NT
NA

RE
S
SORL

S
IE
R
IZ HE

Y
E

R
PE

E
KE

I P
X
AR

N
EY

EA

R
R

T
EN

S
S

N
NH

D
CH
G
TE

E IN
A

AA

S
R
S

UE

T
Y
RD
R

Figure 6. In the timeline view, influence edges result in a wave-like form indicating the propagation of philosophical ideas
over many time periods. The more significant a philosopher is the larger and possibly wider is the resulting wave (below).

LEGEND
modest fireworks than the most influential philosophers, such as Kant. Besides the number of Kant
Immanuel edges, the
(IPA: [!'manu"l kant];spatial
22 April 1724 –
12 February 1804) was an 18th-century German philosopher
SIGNIFICANCE
extent of edges conveys the topical scope of a philosopher’s incoming and outgoing influences.
degree of influence
1 from the Prussian city of Königsberg (now Kaliningrad,
Russia). He is regarded as one of the most influential
47
thinkers of modern Europe and of the late Enlightenment.
In the temporal layout, the influence edges lead to distinct wave-like patterns (see Figure 6). The resulting
Kant created a new widespread perspective in philosophy
which influenced philosophy through the 21st Century. He
INFLUENCE B A C

edge layout is particularly interesting as the waveform can also be seen as a representation ofalso
B influenced A, and B was influenced by C works
published important works of epistemology, as well as
the constant
relevant to religion, law, andcritique
history....

and evolution of philosophical ideas. While in the interest map the spatial extent of edges stands for topical
scope, in the timeline the extent of edges represents the temporal scope of influence.

5.5 Visual Presentation

In addition to the conceptual representation of multiple types of data relationships, it is important to consider
how the visual presentation can combine these with the display of additional information, such as the names of
philosophers, their biographical backgrounds, and visualization-related legends.
To reduce visual clutter, only the names of those philosophers that are currently active or associated with the
active philosopher are displayed. The nodes for the remaining philosophers are dimmed and do not have a label.
 This allows the viewer to focus on the current selection, but it also alleviates a label-occlusion problem that
still occurs when many edges of associated nodes are displayed (see Figure 5, right). In the case of overlapping
labels, it is possible to hover with the mouse over the corresponding nodes to make hidden labels visible. To
ensure aesthetic proportions between circles and labels, the font size is set relative to the size of the circles.

AR TRE

M
W RS

BWU

LU S GE

HSA

A D PP ER

AY UEVA

RA RAEW N

FC
PO AM S
S NAG EIEGR
MPIRREX

BNR
G AEDT

CR TE OL

M EBIE

CAEIDTG

G EAKU

BEER

SEER OUIR-P
RI L
DA OAERS

H
ST S

LY LS ON
DEOT
PULE RD
HBURDOESYC

H IT
EM UKV

SAORER
W

OHTNUZ
Furthermore, only the surname of the philosopher is used, which is typically unique and sufficiently known.

TYR

AR O
O LZ

AD S
HI GESR
AC LS

IS NR
ES N HE

CL
Ká

EN
I

W S

UOCA E
AL
RNEGEN

VI UR
O HE N

R NO
E IN
R O

CE A

MAU
D
CS

DT

A
AP G S

SKL
M
To learn more about an individual philosopher, a detailed view is displayed in the lower right portion of the

YT
R
D

ERTE
AA R

O
window (see Figure 7, left). The beginning of the corresponding Wikipedia article is displayed and accompanied

NT
A UE

RD

Y
IN
with a visual depiction (e.g., photo, painting, sculpture) of the philosopher. Clicking anywhere on the detail
window opens the corresponding entry on Wikipedia. The border of this background display uses the colour of
the corresponding philosopher.

Immanuel Kant (IPA: [!'manu"l kant]; 22 April 1724 –

12 February 1804) was an 18th-century German philosopher
from the Prussian city of Königsberg (now Kaliningrad,
Russia). He is regarded as one of the most influential
thinkers of modern Europe and of the late Enlightenment.
Kant created a new widespread perspective in philosophy
which influenced philosophy through the 21st Century. He
also published important works of epistemology, as well as
works relevant to religion, law, and history....

Figure 7. Detailed information for currently activated philosopher (left). Legend for node sizes and edge types (right).

Legends are displayed to summarize the different mappings that are used. For example, for the size of circles
and types of edges, a legend for both layouts is always displayed in the lower left of the screen (see Figure 7, right).
The two circle sizes displayed actually correspond to the smallest and largest nodes in the visualization, giving
a sense for the extent of significance between philosophers. In the timeline view, years indicate the temporal
distribution along the time axis. The legends are drawn in shades of light grey to avoid distracting from the
information visualization.

6. EXPLORING COMPLEX RELATIONSHIPS

Allowing only one philosopher to be selected at a time can be seen as a limitation, but as suggested before it
opens up interesting possibilities for visualizing different patterns and relationships. Furthermore, having only
one philosopher selected at a time has interesting implications for the interaction with NLDs and MDS-based
visualizations. Selecting a philosopher draws the incoming and outgoing edges, highlights the respective nodes,
and displays their names. As these linked nodes are revealed, the viewer is more likely to follow the displayed
edges and activate one of the linked philosophers. In a sense, exploring philosophers along their influence edges
becomes a pivoting exploration from one philosopher to the next. It would be interesting to examine whether the
structure of networks is easier to comprehend by actively pivoting through it or by looking at a global overview
with possibly many overlapping edges.
Consider starting out with the interest map layout and selecting, for example, the node of Beauvoir, which
is one of the smaller circles in the periphery of the visualization (see Figure 8, top). After having selected this
philosopher it is now possible to follow one of the philosophers that was influenced by her. In this case one could
select Deleuze (bottom right) and afterwards a philosopher with a larger and more saturated node, in this case
Nietzsche (bottom left). The path of exploration depends somewhat on serendipity, intuition, and interest, all
of which are affected by the overall node layout, the edges, and the colourings.
To change the layout, a button in the upper left corner triggers the transition between the interest map and
the timeline. As the colours are based on interest similarity it is possible to rediscover nodes from the interest
map in the timeline view. The transition between the views is animated to allow the viewer to follow nodes
between the views.
 SARTRE

KIERKEGAARD

BEAUVOIR BEAUVOIR

CAMUS KANT
HEGEL

MARX

! TIME LINE
DELEUZE DESCARTES
SARTRE SARTRE
HUSSERL

NIETZSCHE HEIDEGGER

STRAUSS

Simone de Beauvoir (pronounced [sim!n d" bo#vwa$

LEGEND French) (January 9, 1908 – April 14, 1986) was a Fre
author and philosopher. She wrote novels, monograp
SIGNIFICANCE philosophy, politics, and social issues, essays,
1
degree of influence
1 biographies, and an autobiography in several volume
47 BEAUVOIR 47 BEAUVOIR now best known for her metaphysical novels, includin
Came to Stay and The Mandarins, and for her 1949
B C FOUCAULT INFLUENCE A B C FOUCAULT The Second Sex, a detailed analysis of women's opp
was influenced by C B influenced A, and B was influenced by C and a foundational tract of contemporary...
CAMUS KANT KANT
MERLEAU-PONTY MERLEAU-PONTY
MAIMON HEGEL
HERACLITUS
SPINOZA SPINOZA
CAPUTO

GORDON
SCHELER
SCHOPENHAUER
LéVINAS SPIR

RAWLS BLANCHOT BENJAMIN BLANCHOT

GUATTARI GUATTARI
PARMENIDES BERGSON

DERRIDA DELEUZE DELEUZE

PLATO
BUBER

RORTY
EMPEDOCLES
NIETZSCHE HEIDEGGER NIETZSCHE
BADIOU
WITTGENSTEIN
PASCAL LéVI-STRAUSS
VATTIMO
ADORNO
WHITEHEAD

Figure 8. Exploratory pivoting along philosophers’ influences from Beauvoir over Deleuze to Nietzsche.
Friedrich Wilhelm Nietzsche (October 15, 1844 – August Gilles Deleuze (French pronunciation: [!il d"løz]), (1
LEGEND 25, 1900) (German pronunciation: [!f"i#d"$ç January 1925 – 4 November 1995) was a French ph
!v$lh%lm !ni#t&%]) was a nineteenth-century German of the late 20th century. From the early 1960s until h
7. CURRENT LIMITATIONS AND FUTURE WORK
SIGNIFICANCE philosopher and classical philologist. He wrote critical death, Deleuze wrote many influential works on philo
degree of influence texts on religion, morality, contemporary culture, literature, film, and fine art. His most popular books w
1 1
47 47
philosophy, and science, using a distinctive German language the two volumes of Capitalism and Schizophrenia:
The presented design and implementation of EdgeMaps are first steps towards integrating explicit and implicit
style and displaying a fondness for metaphor and aphorism. Anti-Oedipus (1972) and A Thousand Plateaus (1980
B C INFLUENCE A Nietzsche's B influenceCremains substantial within and beyond co-written with Félix Guattari. His books Difference a
was influenced by C
data relationships. While the idea of integrating NLD and MDS visualizations is promising, there are several
B influenced A, and Bphilosophy, notably
was influenced by C in existentialism and postmodernism.... Repetition (1968) and The Logic of Sense...
limitations associated with our particular design and the general approach.
We have argued that displaying edges for only one node, i. e., philosopher, at a time allows for novel interaction
and visualization methods and solves some of the edge congestion problems of larger NLDs. However, it would
be interesting to see how techniques addressing edge congestion such as EdgeLens6 and Edge Bundles7 could be
integrated with the idea of combining a layout representing implicit relations with edges for explicit relations.
The differentiation between incoming and outgoing edges led to novel visual patterns in the time and interest
views; it would be interesting to consider other partitioning of edges and nodes to expose different data patterns.
While we used the MDS algorithm in the R environment, it would be beneficial to explore its parameters and,
for example, consider planes with arbitrary rectangular shapes besides squares. The difficulty to interpret the
meaning of position and proximity may be alleviated with the display of edges, yet, it needs further exploration
to find other techniques that make the output of MDS algorithms more accessible and meaningful. One of the
ideas that came up during this work is to label regions in the MDS plane based on representative keywords that
are more common among nodes that are positioned closer to each other. Besides the aim of making the MDS
layout more comprehensible, it could be useful to create a flexible MDS algorithm allowing the viewer to change
how items are positioned. The great challenge for this would be to make this algorithm interactive.
Another open question that needs further exploration is the management of large numbers of overlapping
nodes. In the timeline view, it would be possible to position the nodes vertically, for example, ordered by
significance, i. e., size. The interest map currently places smaller nodes around larger nodes with the same
position. In order to support large numbers of overlapping nodes, it would be possible to indicate the spatial
togetherness using visual bubbles such as the one shown by the bubble sets technique.17

8. CONCLUSION
With the design and realization of EdgeMaps we have explored the visualization of explicit and implicit relations
on an example dataset of philosophers. The interactive visualization tool that we have designed and developed
represents implicit relations (time and interests) as the underlying layouts for node-link diagrams in which nodes
stand for philosophers and edges for explicit influence connections between them. By constraining the selection to
one philosopher at a time, it is possible to visually distinguish between incoming and outgoing edges particularly
by their curvatures. Restricting selections to one philosopher at a time and visually distinguishing influence edges
led to novel visual patterns that resembled the aesthetics of fireworks and waves. In contrast to the notorious
yarn ball effect of some graph visualizations, we have suggested that an interactive pivotal exploration along
edges between nodes may better allow viewers to grasp network structure than complex overviews.

ACKNOWLEDGEMENTS
We thank our colleagues in the iLab at the University of Calgary for valuable advice and assistance, in particular,
Uta Hinrichs for her feedback on drafts of this paper. We are also grateful to the referees for their constructive
feedback and suggestions. Funding was provided by SMART Technologies, NSERC, iCORE, and NECTAR.

REFERENCES
[1] Shneiderman, B., “The eyes have it: A task by data type taxonomy for information visualizations,” in
[Proceedings of the IEEE Symposium on Visual Languages], 336–343 (1996).
[2] Bertin, J., [Semiology of Graphics: Diagrams, Networks, Maps], University of Wisconsin Press (1983).
[3] Stone, M. C., [A Field Guide to Digital Color], ch. 12 Color in Information Display, 277–300, AK Peters
(2003).
[4] Herman, I., Melançon, G., and Marshall, M., “Graph visualization and navigation in information visualiza-
tion: A survey,” TVCG: Transactions on Visualization and Computer Graphics 6(1), 24–43 (2000).
[5] Purchase, H. C., “Metrics for graph drawing aesthetics,” Journal of Visual Languages & Computing 13,
501–516 (October 2002).
[6] Wong, N., Carpendale, S., and Greenberg, S., “EdgeLens: An interactive method for managing edge con-
gestion in graphs,” in [InfoVis 2003: Symposium on Information Visualization], 51–58, IEEE Computer
Society (2003).
[7] Holten, D., “Hierarchical edge bundles: Visualization of adjacency relations in hierarchical data,” TVCG:
Transactions on Visualization and Computer Graphics 12(5), 741–748 (2006).
[8] Becker, R., Eick, S., et al., “Visualizing network data,” TVCG: Transactions on Visualization and Computer
Graphics 1(1), 16–21 (1995).
[9] Heer, J. and danah boyd, “Vizster: Visualizing online social networks,” in [InfoVis 2005: Symposium on
Information Visualization], 32–39, IEEE Computer Society (2005).
[10] Wise, J. A., Thomas, J. J., Pennock, K., Lantrip, D., Pottier, M., Schur, A., and Crow, V., “Visualizing
the non-visual: Spatial analysis and interaction with information from text documents,” in [InfoVis 1995:
Symposium on Information Visualization], 51–58, IEEE Computer Society (1995).
[11] Williams, M. and Munzner, T., “Steerable, progressive multidimensional scaling,” in [InfoVis 2004: Sympo-
sium on Information Visualization], 57–64, IEEE Computer Society (2004).
[12] Wattenberg, M., Res, I., and Cambridge, M., “Arc diagrams: visualizing structure in strings,” in [Info-
Vis 2002: Symposium on Information Visualization ], 110–116, IEEE Computer Society (2002).
[13] Kerr, B., “Thread Arcs: An email thread visualization,” in [InfoVis 2003: Symposium on Information
Visualization], 211–218, IEEE Computer Society (2003).
[14] Neumann, P., Schlechtweg, S., and Carpendale, S., “ArcTrees: visualizing relations in hierarchical data,” in
[Eurographics-IEEE VGTC Symposium on Visualization ], 53–60 (2005).
[15] Shneiderman, B. and Aris, A., “Network visualization by semantic substrates,” TVCG: Transactions on
Visualization and Computer Graphics 12, 733–740 (Sep/Oct 2006).
[16] Collins, C. and Carpendale, S., “VisLink: Revealing relationships amongst visualizations,” TVCG: Trans-
actions on Visualization and Computer Graphics 13, 1192–1199 (Nov/Dec 2007).
[17] Collins, C., Penn, G., and Carpendale, S., “Bubble sets: Revealing set relations with isocontours over
existing visualizations,” TVCG: Transactions on Visualization and Computer Graphics 15(6), 1009–1016
(2009).