Overlap community detection using spectral algorithm based on node convergence degree

Abstract

Community structure is a typical feature of complex networks in cyberspace, and community detection is considered to be crucial to understanding the topology structure, network function and social dynamics of cyberspace. However, some particular nodes may simultaneously belong to several communities in cyberspace. Though there are many algorithms to detect the overlapping communities, most of them are based on the network structure without considering the attributes of the nodes. In this paper, we focus on the convergence characteristic of network and propose an overlap community detection algorithm based on the node convergence degree, which is defined as a combination of attribute convergence degree and structure convergence degree. It combines the network topology with the attributes of the nodes and considers both local and global information of a node. An improved PageRank algorithm is used to get the importance of each node in the global network, while the information of local network is used to measure the structure convergence degree. The overlap communities are thus identified by spectral cluster based on the node convergence degree. Finally, experiment results demonstrate the effectiveness and better performance of our proposed method.