Probabilistic Community Using Link and Content for Social Networks

Abstract

Community detection is one of the most important problems in social network analysis in the context of the structure of underlying graphs. Many researchers have proposed methods, which only consider the network structure of social networks, for discovering dense regions in social networks. However, increasing media information in networks, such as images, videos, user tags, and comments, are observed with the development and application of Web 2.0. Abundant content information is available to provide a different view for community detection process. In this paper, we propose an overlapping community detection method, namely, latent Dirichlet allocation-based link partition (LBLP), which uses a graphical model and considers network structure and content information. Two feature integration strategies are proposed to combine the influence of network structure and content information on the network generation process. Experimental results on synthetic and real-world networks show that the LBLP method is effective, and content information is beneficial in mining community structure.