Community Detection For Micro-blog Network Based On MMSB Model

Community detection mainly based on node link denseness in communities and link sparseness characteristics between communities, and with the help of the network average path length, clustering coefficient and degree distribution of depicting the network structure of attributes to differentiate community. These methods can’t find overlapping community and its computational complexity will become larger because of the increasing of the network nodes. As a result of the high number of the network nodes and the complex relationship between the nodes, the traditional method has limitations for Micro-blog network. MMSB model is a kind of based on Bayesian network, and allows nodes belong to multiple models of the community. Although the model can deal with large-scale network nodes, and the computing complexity is relatively low, but the model is suitable for undirected network.Based on the MMSB model and combined the characteristics of Micro-blog, by improving the old model, I put forward a new community detection model WB-MMSB for Micro-blog. Considering the nodes of Micro-blog Network have single direction relations, WB-MMSB is put forward for community detection, which uses directed edges to embody the direction relations of nodes, from two aspects link-in and link-out to quantify the community membership of nodes, uses exponential family distribution and mean-field variational inference method to inference the representations of variables in this model, and uses SVI algorithm to compute relating parameters. Experiments adopt Sina-Weibo dataset and NMI to testify the performance of WB-MMSB. The results indicate that the community detection ability of WB-MMSB model is better than aMMSB model, the convergence rate of WB MMSB model is faster than aMMSB model.