TOWARDS USER-CENTRIC SOCIAL NETWORKS

Panayiotis Andreou

, Panagiotis Germanakos

, Andreas Konstantinidis

Dimosthenis Georgiadis

, Marios Belk

and George Samaras

Department of Computer Science, University of Cyprus, P.O. Box 20537, 1678 Nicosia, Cyprus

Institute of Informatics and Telematics - CNR, 56124, Pisa, Italy

Department of Computer Science, Frederick University, 1036 Nicosia, Cyprus

Department of Management and Management Information Systems, University of Nicosia, Nicosia, 1700, Cyprus

Keywords:

Smartphone Networks, Social Networks, Multi-objective Optimization, Decision Making, Adaptivity.

Abstract:

Social network portals, such as Facebook and Twitter, often discover and deliver relevant social data to a

user’s query, considering only system-oriented conﬂicting objectives (e.g., time, energy, recall) and frequently

ignoring the satisfaction of the individual “needs” of the query user w.r.t. its perceptual preference characteris-

tics (e.g., data comprehensibility, working memory). In this paper, we introduce User-centric Social Network

(USN), a novel framework that deals with the conﬂicting system-oriented objectives of the social network

in the context of Multi-Objective Optimization and utilizes user-oriented objectives in the query dissemina-

tion/acquisition process to facilitate decision making. We present the initial design of the USN framework and

its major components. Our preliminary evaluation with real datasets shows that USN enhances the usability

and satisfaction of the user while in parallel provides optimal system-choices for network performance.

1 INTRODUCTION

The evolution of smartphone devices along with the

ascend of social networks has enabled the invention

of myriad of applications that allow users to continu-

ously interact and share social data. This data is typ-

ically accessed using a portal provided by the social

network provider, which enables querying the social

data based on keywords that describe their content.

It is a fact that the environment of most social net-

work portals is not user-centric (i.e., social content is

presented using a global representation scheme appli-

cable to all users). However, this global representa-

tion scheme is not always optimized based on spe-

ciﬁc user intrinsic characteristics (e.g., working mem-

ory span). In order to address the comprehension and

orientation difﬁculties presented in such systems and

satisfy the heterogeneous needs of the users, a num-

ber of researchers studied adaptivity and personaliza-

tion (Brusilovsky, 2001; Lankhorst et al., 2002; Ger-

manakos et al., 2008).

The process of content adaptation takes into ac-

count the parameters included in the user proﬁle (e.g.,

working memory span, cognitive style) and returns

the best adaptive environment that meets the indi-

vidual preferences and demands of each user. How-

ever, enabling dynamic adaptation of the environment

while in parallel aiming to optimize the runtime per-

formance requirements of the network is not a trivial

task as it requires tackling with a number of conﬂict-

ing parameters (e.g., energy, time, usability). Because

so many different parameters are involved, the respec-

tive problem is a proper object for Multi-objective Op-

timization (MOO). In MOO, there is no single solu-

tion that optimizes all objectives simultaneously but

instead a set of non-dominated solutions commonly

known as the Pareto Front (PF). Our framework opts

for a subset of these solutions that increase the us-

ability of the social network taking into account the

individual preferences of each user, facilitating in this

way decision making.

In particular, in this paper we present User-centric

Social Network (USN), a novel framework that com-

bines system-oriented with user-oriented objectives in

order to increase both the network performance as

well as the query user’s satisfaction. To the best of

our knowledge, no previous work has combined the

disciplines of multi-objective optimization and deci-

sion making with content adaptation and personaliza-

tion in order to increase both network performance as

well as usability.

795

Andreou P., Germanakos P., Konstantinidis A., Georgiadis D., Belk M. and Samaras G..

TOWARDS USER-CENTRIC SOCIAL NETWORKS.

DOI: 10.5220/0003932007950798

In Proceedings of the 8th International Conference on Web Information Systems and Technologies (WEBIST-2012), pages 795-798

ISBN: 978-989-8565-08-2

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

2 BACKGROUND AND RELATED

WORK

We now provide related research work on multi-

objective optimization and cognitive user proﬁles that

lie at the foundation of the USN framework.

Multi-Objective Optimization (MOO) & Decision

Making. It has been shown that Multi-Objective

Evolutionary Algorithms (MOEAs) are more effec-

tive in tackling Multi-objective Optimization Prob-

lems (MOPs), as opposed to existing linear/single

objective methods. In the literature, several MOPs

were proposed within the context of Wireless Sen-

sor Networks and Mobile Networks, tackled in most

cases by Pareto-dominance based MOEAs, such

as the state-of-the-art Non-Dominated Sorting Ge-

netic Algorithm-II (NSGA-II) (Deb et al., 2002).

The particular class of decompositional MOEAs

(MOEA/D) (Zhang and Li, 2007) utilized in this

work, have been shown to be efﬁcient and effective

with combinatorial real life MOPs (Konstantinidis

et al., 2010b; Konstantinidis et al., 2010a) by incor-

porating scalar knowledge and techniques. In general,

a MOP solution obtained by MOEA refers to a feasi-

ble set of pareto-optimal solutions without commit-

ting any information about what represents a suitable

compromise solution. This is due to the fact that all

solutions are equally important. Therefore, in most

cases a decision making phase (Chaudhuri and Deb,

2010) is required after the optimization phase to ad-

dress this problem (i.e., select the most suitable com-

promise solution from the pareto-optimal set). A de-

cision maker (Chaudhuri and Deb, 2010) is usually a

human expert about the problem and is utilized for de-

ciding which is the most appropriate solution. In our

setting, the decision making is accomplished using

the user-oriented objectives derived from the query

user’s cognitive proﬁle.

Cognitive User Proﬁles. Effective personalization of

content involves two important challenges: i) accu-

rately identifying users comprehensive proﬁles, and

ii) adapting any content and processes in such a

way that enables efﬁcient and effective navigation

and presentation to the user. User Perceptual Pref-

erence Characteristics (UPPC) (Germanakos et al.,

2008), serve as the primal personalization ﬁltering el-

ement, which apart from the “traditional” (predeter-

mined characteristics), emphasizes on a different set

of characteristics, which inﬂuence the visual, mental

and emotional processes that mediate or manipulate

new information that is received and built upon prior

knowledge, respectively different for each user or user

group. It has been shown in environments such as

Device Level

(Smartphone Users)

Server Level

(Social Network)

Optimization

Phase

Decision

making Phase

“Find data

about

Parthenon in

Athens”

Pareto

Front

PF={x

,...}

Result

“X images and

Y articles about

Parthenon”

Social

Data

System-oriented

Objectives

{S1, S2, S3}

User-oriented

Objectives

{U1, U2}

User

Profiles

Figure 1: USN framework architecture.

eLearning and eServices (Germanakos et al., 2008)

that these characteristics have a major impact on vi-

sual attention, cognitive and emotional processing.

In our context-based mobile social network set-

ting, we have opted for two representative cognitive

factors (i.e., user-oriented objectives), the Cognitive

Style and Working Memory Span that are consid-

ered of high signiﬁcance in such environments (Ger-

manakos et al., 2008; Graf and Kinshuk, 2009).

Mainly, our approach has been driven by the differ-

ence in cognitive information processing capabilities

of the user.

3 USN FRAMEWORK

In this section, we provide the architecture of the USN

frameworkincluding descriptions of its major compo-

nents. Figure 1 illustrates the components of the USN

framework and their interactions.

In the USN framework, each smartphone device

stores its data (e.g., images, documents) in the de-

vice’s local storage. When a user u

decides to

search for social data, the device’s interface gener-

ates a query Q and disseminates it to the social net-

work. The social network portal recursively forwards

Q to users not in close location or social proximity

to u

, similar to (Andreou et al., 2011). As soon as

candidate users are selected (i.e., users that can par-

ticipate in Q ) they are forwarded to the Optimizer

that generates solutions (i.e., sets of users, their so-

cial data and the connectivity among them), which

are then evaluated using the system-oriented objec-

tives until the set of non-dominated solutions (PF) is

generated. The PF is then fed to the Decision Maker,

which takes as input the query user’s proﬁle and ex-

tracts the user-oriented objectives. Each solution in

PF is then ranked using the ﬁtness error (calculated

by the user-oriented objectives and the values in the

query user’s proﬁle). The data of the most efﬁcient

WEBIST2012-8thInternationalConferenceonWebInformationSystemsandTechnologies

796

solution are returned to the query user’s smartphone.

We nowprovidemore detailed information on the ma-

jor components of the USN framework.

3.1 User Proﬁles

The User Proﬁle comprises of all the information re-

lated to the user such as traditional characteristics

(e.g., age) and cognitive characteristics (e.g., work-

ing memory span). Additionally, each user proﬁle

is dynamically updated by continuously proﬁling the

volatile characteristics of the user (e.g., time and lo-

cation, navigation experience).

3.2 Optimizer

The USN optimizer utilizes the MOEA/D approach

for generating the Pareto-optimal set of solutions (i.e.,

Pareto-Front). In order to accomplish this, the MOP

is ﬁrstly decomposed into m subproblems (Zhang and

Li, 2007). The i

subproblem is in the form

maximize g

(G |w

, z

∗

) = max{w

| f

(G ) − z

∗

|} (1)

where G denotes the Social Network Graph and f

(j = S1, S2, S3) are the system-oriented objectives of

our MOP, which are described below:

S1: Minimize the total Energy consumption of G

Energy(G ) = MIN(

∑

∈G

e(u

, Q )). (2)

where, e(u

, Q) denotes the energy consumption for

transmitting all data objects of u

that satisfy the ﬁlters

of Q over the respective edge (WiFi, Bluetooth, 3G).

S2: Minimize the Time overhead of G

Time(G ) = MIN(

∑

∈G

t(u

, Q)). (3)

where, t(u

, Q) denotes the time overhead for trans-

mitting all data objects of u

that satisfy the ﬁlters of

Q over the respective edge.

S3: Maximize the Recall rate of G

Recall(G , Q ) = MAX(

Relev.(G , Q ) ∩Retriev.(G , Q )

Relev.(G , Q )

)

(4)

Our framework utilizes the aforementioned sys-

tem objectives in order to obtain the pareto-front PF.

In the ﬁnal step, the generated PF solutions are fed

into the Decision Maker for ranking.

3.3 Decision Maker

In order to facilitate decision making and opt for the

most user-efﬁcient solutions, the Pareto-optimal so-

lutions X ∈ PF obtained are then evaluated using

U1:Comprehension Ability and U2:Cognitive Over-

load user-oriented objectives. Note that the values for

U1 and U2 are extracted from the proﬁle p

of user u

U1: Maximize Comprehension Ability

CA(X , p

) = MAXcs(r(X ), p

). (5)

where, cs(r, p

) denotes the evaluation of the compre-

hension ability of user u

over the results r(X ) based

on its cognitive style.

U2: Minimize Cognitive Overload:

CO(X , p

) = MIN(wm(r(X ), p

)). (6)

where, wm(r, p

) denotes the evaluation of the cogni-

tive overload of user u

over the results r(X ) based on

its working memory.

Decision Making/Support Fitness Error

In order to rank each PF solution, we deﬁne the ﬁtness

error as the distance of a solution X from the opti-

mal solution (i.e., the difference between the obtained

user-oriented objective values and the actual/exact

values provided from the user proﬁle).

FitnessError = |CA(X , p

) − p

| + |CO(X , p

) − p

(7)

In the ﬁnal step, USN ranks the solutions based on

the ﬁtness error and returns either the ﬁrst one (i.e., au-

tomated decision making) or the k-most important ones

(i.e., decision support). As soon as the ﬁnal set of solu-

tion(s) is produced, the Decision Maker returns the re-

sults to the query processing mechanism, which in turn

forwards the results to the query user.

4 EXPERIMENTAL EVALUATION

We have performed a preliminary evaluation of the

USN framework using real datasets. We obtained user

proﬁles from the AdaptiveWeb project (http:// adap-

tiveweb.cs.ucy.ac.cy/), which includes user proﬁles of

327 students of the University of Cyprus and University

of Athens; 40% male, and 60% female, with ages

varying from 19 to 23. Each proﬁle contains informa-

tion regarding the student’s cognitive characteristics

including his/hers Cognitive Style (objective U1) and

Working Memory Span (objective U2). These proﬁles

were derived after running a number of psychometric

experiments provided by the AdaptiveWeb Project.

Additionally, each user proﬁle from the UPPC dataset

TOWARDSUSER-CENTRICSOCIALNETWORKS

797

0.4

0.6

0.8

1.2

1.4

1.6

1.8

2.2

2.4

100

Recall (%)

Comparison of USN solutions

Pareto Front

Top-5

Best

Energy (mJx10

-3

)

Time (s)

Recall (%)

Figure 2: Optimal and Top-k solutions compared to the

Pareto-Front (PF) solutions provided by USN.

was augmented with the user’s social data content of

Facebook. Finally, in order to introduce mobility in

our experiments, we have utilized a publicly available

real dataset by Microsoft Research Asia GeoLife,

which includes 1,100 trajectories of a human moving

in the city of Beijing over a life span of two years

(2007-2009). At each timestamp, we select a user u

as the query user and execute the following query (in

SQL-syntax:

= ``SELECT * FROM Users WHERE keyword

filter

, where ﬁlter is a keyword.

We study the Pareto-Front (PF) solutions provided

by the USN framework. More speciﬁcally, we compare

the ﬁtness error provided by the best solution and the

top-k solutions. In Figure 2, we demonstrate the results

for a single timestamp (τ=19) for all solutions in the

system-oriented objective space with the Energy,Time

and Recall metrics. The PF solutions are represented by

solid circles. The Top-k (k=5) solutions and the best so-

lution are represented by diamonds and a solid triangle,

respectively.

We observe that the Top-k solutions, w.r.t. the ﬁtness

error provided by the USN framework, almost spread

across the whole system-oriented objective space. This

is important as it enables the network decision maker

to efﬁciently tune the system according to speciﬁc net-

work requirements (e.g., low energy is more impor-

tant than low time and high recall objectives) provid-

ing at the same time near-optimal user-oriented ﬁtness.

The execution time required for generating the solutions

was ≈32562±3409ms, which is not applicable for sys-

tems requiring realtime performance. However, cloud-

computing or parallel processing can alleviate this prob-

lem by evaluating each solution in each generation in-

dependently. Since network operators typically employ

server farms that feature thousands of processing cores

running in parallel, the execution time can be reduced

by several orders of magnitude thus offering realtime

performance.

5 CONCLUSIONS

In this paper, we introduced User-centric Social Net-

work (USN), a novel framework that incorporates user-

oriented objectives in the search process. We presented

the initial design of the USN framework as well as a

preliminary evaluation, which demonstrates that USN

enhances usability and satisfaction while in parallel op-

timizing the performance of the network w.r.t. energy,

time and recall.

ACKNOWLEDGEMENTS

This work is partly supported by the European Union

under the project CONET (#224053), and the project

FireWatch (#0609-BIE/09), sponsored by the Cyprus

Research Promotion foundation.

REFERENCES

Andreou, P., Zeinalipour-Yazti, D., Pamboris, A., Chrysan-

this, P., and Samaras, G. (2011). Optimized query

routing trees for wireless sensor networks. Informa-

tion Systems Journal, 36(2):267–291.

Brusilovsky, P. (2001). Adaptive hypermedia. User Model-

ing and User-Adapted Interaction, 11:87–110.

Chaudhuri, S. and Deb, K. (2010). An interactive evolution-

ary multi-objective optimization and decision making

procedure. Applied Soft Computing, 10(2):496–511.

Deb, K., Pratap, A., Agarwal, S., and Meyarivan, T. (2002).

A fast and elitist multiobjective genetic algorithm:

NSGA II. IEEE TEC, 6(2):182–197.

Germanakos, P., Tsianos, N., Lekkas, Z., Mourlas, C., and

Samaras, G. (2008). Realizing comprehensive user

proﬁle as the core element of adaptive and personal-

ized communication environments and systems. The

Computer Journal, 52(7):749–770.

Graf, S. and Kinshuk (2009). Advanced adaptivity in

learning management systems by considering learn-

ing styles. In IEEE/WIC/ACM, volume 3.

Konstantinidis, A., Charalambous, C., Zhou, A., and Zhang,

Q. (2010a). Multi-objective mobile agent-based sen-

sor network routing using MOEA/D. In IEEE CEC.

Konstantinidis, A., Yang, K., Zhang, Q., and Zeinalipour-

Yazti, D. (2010b). A multi-objective evolutionary

algorithm for the deployment and power assignment

problem in wireless sensor networks. Elsevier Com-

puter Networks, 54:960–976.

Lankhorst, M., Kranenburg, A. S., and Peddemors, A.

(2002). Enabling technology for personalizing mobile

services. In HICSS, volume 3, page 87.

Zhang, Q. and Li, H. (2007). MOEA/D: A multi-objective

evolutionary algorithm based on decomposition. IEEE

TECC, 11(6):712–731.

WEBIST2012-8thInternationalConferenceonWebInformationSystemsandTechnologies

798