August 12, 2014 23:4 WSPC/INSTRUCTION FILE CSBioEGCE Detection of Highly Overlapping Communities in Complex Networks Madhusudan Paul 1 , Rishav Anand 2 , and Ashish Anand 3 Department of Computer Science and Engineering Indian Institute of Technology Guwahati, India 1 madhusudan@iitg.ernet.in 2 rishav@alumni.iitg.ernet.in 3 anand.ashish@iitg.ernet.in Detecting communities in complex networks is one of the most important aspects to understand complex systems. In reality, many of these communities are highly overlap- ping in nature, i.e., several nodes belong to more than three communities. Identification of highly overlapping communities are strongly demanded in many applications such as systems biology and social networks. Although there are algorithms for detecting overlap- ping communities, majority of these are unable to detect highly overlapping communities properly. The performance of these algorithms falls sharply when overlapping nodes be- long to more than three communities. In this paper, we propose an extension of existing overlapping community detection algorithm, namely Greedy Clique Expansion (GCE). Due to lack of unavailability of real networks with complete information of ground-truth communities, firstly, we experiment on state-of-the-art synthetic benchmark datasets and observe that our proposed extension exhibits better performance when overlapping nodes belong to more than three communities. We also experiment on real datasets and ob- serve competitive performance. The proposed extension can be applied on networks with highly overlapping community structure such as protein-protein interaction networks. Keywords : Overlapping community detection; Greedy Clique Expansion; complex net- works. 1. Introduction Community structure is an inherent property found in most of real networks such as social networks, biological networks. These communities have significant impor- tance in these networks. For example, persons studied from same school are most likely to form a community in social networks, proteins having similar biological functions are most likely to form a complex (i.e., a community). In many cases, it is found that these communities are overlapping in nature, i.e., a single node may belong to multiple communities. For example, a typical Facebook user be- longs to several distinct communities, similarly, many proteins belong to multiple complexes in protein-protein interaction networks 17,15 . In complex networks the- ory, community detection problem is one of the most fundamental problem. From computational viewpoint, detecting communities in complex networks is computa- tionally challenging task, since many problems related to determining structural properties of graphs are often NP-hard in nature 5 . The complexity of detecting communities rises rapidly as degree of overlapping increases. Although several algo- 1