Community detection in attributed networks using stochastic block models

Community detection is a significant focus in the field of complex network analysis. Most existing community detection methods for attributed networks primarily rely on network structure alone, while approaches that incorporate node attributes are typically designed for traditional community structures. These methods struggle to identify multipartite and mixed structures within the network. In addition, the model-based community detection methods proposed for attributed networks so far have not fully incorporated the distinctive topological information of nodes, such as betweenness centrality and clustering coefficient. In this paper, we propose a stochastic block model that incorporates betweenness centrality and clustering coefficient of nodes for attributed networks, referred to as BCSBM, along with an improved version called PBCSBM. Unlike other generative models for attributed networks, the BCSBM and PBCSBM models generate node links and attributes independently, with both processes following a Poisson distribution. Additionally, the connection probability between communities is determined based on the stochastic block model. Furthermore, the BCSBM model incorporates the betweenness centrality and clustering coefficient of nodes into the generation process for both links and attributes. The PBCSBM model is an improvement over the BCSBM model, taking into account the influence of node degree, betweenness centrality and clustering coefficient on the algorithm's accuracy. Finally, the parameters of the BCSBM and PBCSBM models are inferred using the Expectation-Maximization (EM) algorithm, and the node-community memberships in the network are determined through a hard clustering process. Compared with six algorithms on three attributed networks containing different network structures, the results show that the BCSBM and PBCSBM demonstrate strong data fitting capabilities and better performance.