Node-community membership diversifies community structures: An overlapping community detection algorithm based on local expansion and boundary re-checking

Abstract

Local expansion methods excel in efficiency for mining overlapping communities in real-world networks. However, two problems prevent such methods from identifying diversely structured communities. First, local expansion methods generate independent communities only. Second, local expansion methods depend heavily on quality functions. This work provides a solution for local expansion methods to identify diversely structured communities. The proposed overlapping community detection algorithm performs local expansion and boundary re-checking sub-processes in order. The local expansion process first gets a cover of the network, and then the boundary re-checking process optimizes the cover of the network resulting from the local expansion process. To solve the first problem, the proposed algorithm establishes associations between boundaries of adjacent communities via the boundary re-checking process. To solve the second problem, the proposed algorithm expands and optimizes communities based on node-community membership optimization. We compared the proposed algorithm to seven state-of-the-art algorithms by examining their performance on five groups of artificial networks and sixteen real-world networks. Experimental results showed that the proposed algorithm outperforms compared algorithms in identifying diversely structured communities.