Distributed Clique Percolation based community detection on social networks using MapReduce

In study of complex networks, valuable insights can be obtained by mining structural and functional sub-units of networks, usually called communities, modules, or clusters. One of the approaches to community detection is Clique Percolation Method (CPM) which is the most popular overlapping community detection method and in recent years has been used in analysis of different kinds of networks. However, application of CPM even on small size social networks is very challenging due to its extensive memory, processing, and IO requirements. Hence it is necessary to use distributed and parallel computing models to tackle CPM´s computational challenges. In this paper a new distributed algorithm for computation of CPM will be introduced. The new algorithm is based on the MapReduce distributed computing model which extensively has been used to solve large scale data processing problems. Experimental results will be provided to show that the new MapReduce based algorithm for computation of CPM outperforms the best available algorithms with one order of magnitude when benchmarking them against real-world social network datasets.