Abstract experienced explosive growth. Social websites, such as

Abstract – Detecting overlapping communities in social network is
an important problem in the Data mining. In existence, overlapping community
detection with swarm intelligence frequently generates overlapping community
structures with surplus small communities. To deal with this problem, an
efficient algorithm LEPSO based on line graph, ensemble learning and particle
swarm optimization (PSO) proposed for overlapping community detection. A
discrete PSO, consists of an encoding scheme with ordered neighborhood list and
particle updating tactic with ensemble clustering, is devised for improving the
optimization ability to search the hidden communities in the social networks.
Then, a post processing approach is accessible for merging the finer-grained
and suboptimal overlapping communities. Experiment on some real world datasets
by using the LEPSO is superior in terms of robustness, effectiveness and
automatically determine the number of clusters.

Keywords: Particle Swarm Optimization, Overlapping Community,
Line Graph, Ensemble learning, Social Network Analysis.

 

I. INTRODUCTION

  In the last decade, the social network has
experienced explosive growth. Social websites, such as YouTube and Flickr, have
billions of users sharing their information, users’ opinion and videos every
day. In order to interact users have various features like a comment,
subscribe, like a reply, etc. users easily communicate with each other. The
users are interested to share their information around particular topics they
are represented as communities.

 

Community detection is a great impact on understanding the organization
and the functions of the group are to be detected. Community mining techniques
and theories have developed various applications such as photo tagging, video
gloss etc.

Community
mining techniques and theories have developed various applications such as
photo tagging, video gloss etc. For example 1 involved detecting communities
based on edge betweenness, the counts of short loops in networks and voltage
differences in resistor networks. This method is ideal for the types of
real-world network data with which current research is concerned. M. Cheung,
2 proposed a Genetic Algorithm is used for detecting communities in complex
networks for finding optimizing network. The Algorithm has O (e) time
complexity and does not need prior knowledge about the number of communities.
L. Wang and X.-S. Hua 3 proposed an algorithm for network correlation based
social friend recommendation and correlated different “social role” networks to
find a relationship and make friend recommendations. Networks are aligned by
selecting important features from each network. S. Fortunato 4 proposed to
detect network community a new model based on particle swarm optimization. They
are extremely sensitive to the initial solution, easy to fall into the local
optimum.

A
social network can be represented as a graph. The graph consists of nodes and
edges. Edges can be used to interact with the nodes. Optimizing the graph we
need some objective functions, such as modularity Q 4. The swarm intelligence
algorithms are useful for overlapping communities’ detection. PSO does not
capture fully community structure.

 

 

 

 

 

 

 

 

 

 

Fig 1. Drawbacks
of traditional approaches based on swarm intelligence optimization. (a) Real community
structure. 

(b)
generated community structure (Q = 0.42).