Proceedings of International Conference on Advances in Computer Technology and Management (ICACTM)

In Association with Novateur Publications IJRPET-ISSN No: 2454-7875

ISBN No. 978-81-921768-9- 5
February, 23rd and 24th, 2018
LINKZONE SOCIAL NETWORKING RECOMMENDER SYSTEM
DR. KOTRAPPASIRBI
Professor in Computer Science & Engineering, B.E, M.S, M Tech(CSE), Ph.D(CSE), MISTE
KLE's Dr MSS College of Engg., & Technology, Belagavi, India
kotrappa06@gmail.com

DR. KSHAMA V. KULHALLI

Vice-Principal & HOD IT D.Y. Patil College of Engineering and Technology, Kolhapur-416006
kvkulhalli@gmail.com

MR. ABHIJIT J. PATANKAR

Research Scholar, Computer Science and Engineering, VTU, Belgaum, Karnataka
abhijitpatankar@yahoo.com

ABSTRACT returns a list of people with highestrecommendation

Existing social networking services recommend scores to the query. Finally, Linkzone integrates a
friends to users based on their social graphs, which feedback mechanism to further improve the
may not be the most appropriate to reflect a user’s recommendation accuracy. We have implemented
preferences on friend selection in real life. In this Linkzone on the Android-based smartphones, and
paper, we present Linkzone, a novel semantic-based evaluated its performance on both small scale
friend recommendation system for social networks, experiments and large-scale simulations.
which recommends friends to users based on their The results show that the recommendations accurately
life styles instead of social graphs. By taking reflect the references of users in choosing friends. As
advantage of sensor-rich smartphones, Linkzone time passes, World Wide Web (WWW) goes on growing.
discovers life styles of users from user-centric Lots of information is available on WWW. All the
sensor data, measures the similarity of life styles information which we get is not relevant, only few of
between users, and recommends friends to users if them are relevant. When a user tries to search something
their life styles have high similarity. Inspired by text on WWW she/he lands up with thousands of result. As a
mining, we model a user’s daily life as life result, she/he will mess up with huge information. Hence
documents, from which his/her life styles are fetching the actually required details becomes
extracted by using the Latent Dirichlet Allocation cumbersome and time consuming. This gives rise to data
algorithm. We further propose a similarity metric to filtering system. In earlydays, for data filtering,
measure the similarity of life styles between users, Information Filtering (IF) was used. IF was basically
and calculate users’ impact in terms of life styles developed for filtering documentation, articles, news etc.
with a friend-matching graph. Upon receiving a Looking to our era, e-commerce is growing explosively.
request, Linkzone returns a list of people with Whenever a user makes a search for particular item on
highest recommendation scores to the query user. internet to buy, she/he will get many options. Looking at
Finally, Linkzone integrates a feedback mechanism the options user gets confuse what to buy, and will not
to further improve the recommendation accuracy. able to sort the item that is suitable to him/her.
KEYWORDS: social networking, semantic, This problem gave rise to Recommendation System [RS].
recommendation, Linkzone Arecommender system is a personalization system that
helps users to find items of interest based on their
I. INTRODUCTION preferences. Recommender systems are efficienttools
Existing social networking services recommend friends that overcome the information overload problem by
to users based on their social graphs, which may not be providing users with the most relevant contents. The
the most appropriate to reflect a user’s preferences on importance of contextual information has been
friendselection in real life. In this paper, we present recognized by researchers and practitioners in many
Linkzone, a web services for social network, which disciplines including Ecommerce, personalized IR,
recommends friends to users based on their life styles ubiquitous and mobile computing, data mining,
instead of social graphs. By taking advantage of sensor- marketing and management. There are many existing e-
rich smartphones, Linkzone discovers life styles of users commerce websites which have implemented
from user-centric sensor data, measures the similarity of recommendation systems successfully. We will discuss
life styles between users, and recommends friends to few website in our coming section that provides
users if their life styles have high similarity. Inspired by recommendation. Items are suggested by looking at the
text mining, we model a user’s daily life as life behavior of like-minded-users. Groups are formed of
documents, from which his/her life styles are extracted such users, and items preferred by such groups are
by using the Latent Dirichlet Allocation algorithm. We recommended to the user, whose liking and behavior is
further propose a similarity metric to measure the similar to the group. In our model we have incorporated
similarity of life styles between users, and calculate user preferences obtained from Social Networking Site.
users’ impact in terms of life styles with a friend- Social Networking sites are used intensively from last
matching graph. Upon receiving a request, Linkzone decade. According to the current survey, Social
105 | P a g e
 Proceedings of International Conference on Advances in Computer Technology and Management (ICACTM)
In Association with Novateur Publications IJRPET-ISSN No: 2454-7875
ISBN No. 978-81-921768-9- 5
February, 23rd and 24th, 2018
Networking sites have the largest data set of users. Each 10. So according to the probabilistic topic model, we
social networking site notes/records each and every have
activity of user (like: what user likes? what user is doing?
what is user’s hobby? Etc). Social Networking site will
prove to be largest domain in understanding the user
behavior. One of the best examples of social networking
is FACEBOOK. According to current news FACEBOOK is
trying to develop algorithm, to understand user Let Li = [p(z1|di) , p(z2|di),……., p(zZ|di)] and
behavior. Social Networking sites can help us in getting
important information of users, such as age, gender, Lj = [p(z1|dj), p(z2|dj),……, p(zZ|dj )] are set of
location, language, actives, likes etc. our model takes the life style vectors of user i and user j,
into account these parameters of the user to recommend respectively.
books. Most of the friend suggestions mechanism relies
on pre-existing user relationships to pick friend The similarity of life styles between user i and user j,
candidates. For example, Facebook relies on a social link denoted by S(i, j), is defined
analysis among those who already share common as follows:
friends and recommends symmetrical users as potential
friends. The rules to group people together include: S(i,j) = Sc(i,j).Sd(i,j)
1) Habits or life style
2) Attitudes
3) Tastes
4) Moral standards Where, Sc(i, j) is used to measure the similarity of the life
5) Economic level; and style vectors of users as a whole,
6) People they already know. Sd(i, j) is used to emphasize the similarity of
users on their dominant life styles.
II. MATHEMATICAL MODEL
We adopt the commonly used cosine similarity metric v
Let S is the Whole System Consists: for Sc(i, j), that is,
S= {U, w, z, d, Q, F, L,}.
Where,
1. U is the set of number of users.
U= {u1,u2,........,un}. Sc(i,j) = cos(Li, Lj)
2. Q is the set of query generated from user. Friend-matching graph:
Q= {q1,q2,.........,qn}. It is a weighted undirected graph, G = (V,E,W),
3. F is the set of feedback of users. Where,
F= {f1,f2,......... ,fn}. V = {v1,v2, …..vn} is the set of users and n is the number
4. Let w is the set of activities of users
w= [w1,w2,…….,wW] E = {e(i, j)} is the set of links between users, and
wherewi is the I th activity and W is the total
number
of activities. W : ER is the set of weights of edges.
5. Let z is the set of life styles There is an edge e(i,j) linking user i and user j if and only
z = [z1, z2,……,zZ] if their similarity
wherezi is the I th life style and Z is the total
number of life styles.
S(i,j) >= Sthr
6. Let d is the set of life documents Where, Sthr is the predefined similarity threshold. The
d = [d1, d2,…….dn] weight of that edge is represented by the similarity,
where di is the ith life document and n is the
total number of users.

7. Let p(wi|dk) is the probability of the activity wi in a W(i,j) = S(i,j)

certain life document dk. We use the following representation to convert the
graph into a matrix representation,
8. Let p(wi|zj) is the probability of how much the
activity wi contributes to the life style zj.

9. Let p(zi |dk) is the probability of the life style zj

embedded in the life document dk.

106 | P a g e
 Proceedings of International Conference on Advances in Computer Technology and Management (ICACTM)
In Association with Novateur Publications IJRPET-ISSN No: 2454-7875
ISBN No. 978-81-921768-9- 5
February, 23rd and 24th, 2018
III. SYSTEM ARCHITECTURE To recognize the unique feature of the face, the system
Existing System: recognizes the basic pixels and creates the base feature
 Most of the friend suggestions mechanism relies of the face. Further a similar possible image of the face is
on pre-existing user relationships to pick friend created using this base feature and the image is classified
candidates like friend of friend. under various folders. The image data previously stored
 Facebook relies on a social link analysis among is compared and the contents from the image are
those who already share common friends i.e. detected and conclusions are made and conversation is
mutual friend and recommends symmetrical done accordingly.
users as potential friends. The rules to group Eigen face algorithm has been used to implement face
people together include: recognition. If the face is recognized, user is greeted by
Habits or life style his name. New users are registered by taking a snapshot
Attitudes of their face and storing it for further interactions.
Tastes Human and robot interact with the help of speech
Moral standards recognition and
Economic level; and People they already know. language processing. Human responses from all the
Disadvantages Of Existing System chats are stored and the database keeps growing with
 Existing social networking services recommend every interaction. Sentences are processed to find out its
friends to users based on their social graphs, object and a dictionary made in MySQL is used to store
which may not be the most appropriate to all the words and its related part of speech, example: eat-
reflect a user’s preferences on friend selection in verb, chair-noun, beautiful-adjective, etc. Since the
real life database grows with every interaction, the robot’s
Proposed System accuracy and aptness increases with it.
 A web services for social networks, which The AI perceived human speech in real time. A speech
recommends friends to users based on their life recognition software (Dragon naturally speaking from
styles instead of social graphs. nuance, LLC) was first applied to convert speech into
 We model a user’s daily life as life documents, text. Next, a 10-second temporal window was used to
from which his/her life styles are extracted by define a local context. Spoken utterances within a
using the Latent Dirichlet Allocation algorithm. context were then compared to spot frequent words that
 Similarity metric to measure the similarity of life were further processed in two specific ways. First,
styles between users, and calculate users’ frequent words were selected as candidate words for
Advantages Of Proposed System object names and would be linked with visual input to
 Recommendeds potential friends to users if they compute word-object associations. Second, frequent
share similar life styles. words were added to a word list that the AI maintained
 The feedback mechanism allows us to measure to keep track of those words that the robot heard before.
the satisfaction of users, by providing a user In speech production, the AI would selectively produce
interface that allows the user to rate the friend those words. In a way, this mechanism made the AI like a
list copycat – repeating what it just heard most frequently in
the recent past. For example, if a human teacher
happened to say “hello” to the AI in multiple times, the AI
would say “hello” back to the human teacher. Thus, the
AI learning system was transparent and straightforward
purely driven by statistical regularities in the data
without complicated inference. The goal was to show
how better data from interaction may lead to better
statistical learning.

V. CONCLUSION AND FUTURE SCOPE

In this paper, we designed the Linkzone, a web services
for social network. It is Different from the other friend
recommendation mechanisms relying on social graphs in
existing social networking services, where as this
Friendbook extracted life styles from user-centric data
Fig 2. System Architecture collected from sensors on the Smartphone and
recommended potential friends to users if they share
IV. PROPOSED WORK similar life styles. We implemented Friendbook on the
The proposed system contains the following modules: Android-based Smartphone’s, and evaluated its
1. Face recognition performance on both small scale experiments and large-
2. Language Processing and Storage scale simulations. The results showed that the
recommendations accurately reflect the preferences of
users in choosing friends. Beyond the current prototype,

107 | P a g e
 Proceedings of International Conference on Advances in Computer Technology and Management (ICACTM)
In Association with Novateur Publications IJRPET-ISSN No: 2454-7875
ISBN No. 978-81-921768-9- 5
February, 23rd and 24th, 2018
the future work can be four-fold. First, we would like to 4) L. Bian and H. Holtzman.Online friend
evaluate our system on large-scale field experiments. recommendation through personality matching and
Second, we intend to implement the life style extraction collaborative filtering. Proc. of UBICOMM, pages
using LDA and the iterative matrix-vector multiplication 230-235, 2011.
method in user impact ranking incrementally, so that 5) C. M. Bishop. Pattern recognition and machine
Friendbook would be scalable to large-scale systems. learning.Springer New York, 2006.
Third, the similarity threshold used for the friend- 6) Eunjoon Cho, Seth A. Myers, and Jure Leskovec.
matching graph is fixed in our current prototype of Friendship and mobility:user movement in
Friendbook. It would be interesting to explore the location-based social networks. In Proceedingsof
adaption of the threshold for each edge and see whether the 17th ACM SIGKDD International Conference on
it can better represent the similarity relationship on the KnowledgeDiscovery and Data Mining, San Diego,
friend-matching graph. At last, we plan to incorporate CA, USA, August 21-24, 2011,pages 1082–1090,
more sensors on the mobile phones into the system and 2011.
also utilize the information from wearable equipments 7) Jeffrey Fekete. Making the Big Game: Tales of an
(e.g., Fitbit, iwatch, Google glass, Nike+, and Galaxy Gear) Accidental Spectator.page 142, 2013.
to discover more interesting and meaningful life styles. 8) Xiang Ji. Social data integration and analytics for
health intelligence.In VLDB PhD Workshop, 2014.
REFERENCES 9) V. Levenshtein. Binary Codes Capable of Correcting
1) G. R. Arce. Nonlinear Signal Processing: A Statistical Deletions, Insertions, and Reversals. Soviet Physics-
Approach. John Wiley & Sons, 2005. Doklady, 10(8):707–710, 1966.
2) B. Bahmani, A. Chowdhury, and A. Goel.Fast 10) Xingjie Liu, Qi He, YuanyuanTian, Wang-Chien Lee,
incremental and personalized pagerank. Proc. of John McPherson, andJiawei Han. Event-based social
VLDB Endowment, volume 4, pages 173-184, 2010. networks: linking the online and offline social
3) J. Biagioni, T. Gerlich, T. Merrifield, and J. worlds. In The 18th ACM SIGKDD International
Eriksson.EasyTracker: Automatic Transit Tracking, Conference on Knowledge Discovery and Data
Mapping, and Arrival Time Prediction Using Mining, KDD ’12, Beijing, China, August 12-16,
Smartphones. Proc. of SenSys, pages 68-81, 2011. 2012, pages 1032–1040, 2012.

108 | P a g e