The greedy coupled-seeds expansion method for the overlapping community detection in social networks

Community identification refers to the process of selecting dense clusters with sparse connections to the rest of the graph. These communities naturally overlap in many social networks, while each node can participate in several communities. Revealing these structures represents an important task in network analysis. In fact, it allows to understand the features of networks. However, many community detection approaches fail to provide communities with high quality and better ground truth correspondence in reasonable execution time. The key idea of this paper is to propose a coupled-seed expansion method for the overlapping community detection. Specifically, we construct a coupled-seed by choosing a node and its most similar neighbor and then expand this coupled-seed using a fitness function that improves the identification of local communities. The overlapping modularity, the F-score and the extended normalized mutual information measures are used to evaluate the proposed algorithm. The experimental results on 10 instances of real networks and four sets of LFR networks prove that the proposed method is effective and outperforms the existing algorithms (BigClam, OSLOM, SE, Demon, UMSTMO, LC, Ego-Splitting).