Scaling overlapping community detection algorithms

Community structure is observed in many real-world networks in fields ranging from social networking to biological networks. Over the last decade many approaches have been proposed to efficiently detect the underlying structure of communities in graphs with a greater degree of correctness. This specific area of graph mining is known as community detection. It is one of the most critical components of graph mining. It helps in understanding the underlying properties of the graph. While a lot of effort has been expended on detecting standard partitions in a graph, real world applications are complicated and they exhibit nodes belonging to multiple communities concurrently. Although many algorithms such as Clique Percolation Method, COPRA, etc., have been developed to detect overlapping communities in graphs, they rely on expensive and complicated optimization techniques and do not scale well to real world datasets containing millions of nodes and tens of millions of edges. In this thesis, we propose a vertex centric approach to the problem and we evaluate scalability of overlapping community detection algorithms when implemented using vertex centric graph processing frameworks such as GraphLab/GraphChi. In particular we implemented BigCLAM in GraphChi and were able to achieve speedups of up to 6.5x on large scale real world graphs, thus proving that our approach can give linear speedups on the same hardware settings