A Novel Similarity Measurement for Community Structure Detection

How to identify community structure is a fundamental problem for analysis of complex network. In this paper we propose a novel similarity matrix of the nodes for this purpose, which combines the information of adjacency matrix and common-neighbors matrix. We compare it with diffusion kernel similarity and adjacency matrix using several algorithms which are widely used in detecting community structure, including the standard nonnegative matrix factorization, symmetric nonnegative matrix factorization and spectral clustering. The results performed on the synthetic benchmark networks show that the novel similarity matrix is relatively effective to find the community structures in networks with heterogeneous distribution of node degrees and community sizes, and this effectiveness is also manifested on the real world networks.