Dynamic Stochastic Block Model with Scale-Free Characteristic for Temporal Complex Networks

Complex network analysis has been widely applied in various fields such as social system, information system, and biological system. As the most popular model for analyzing complex network, Stochastic Block Model can perform network reconstruction, community detection, link prediction, anomaly detection, and other tasks. However, for the dynamic complex networks which are always modeling as a series of snapshot networks, the existing works for dynamic networks analysis which are based on the stochastic block model always analyze the evolution of dynamic networks by introducing probability transition matrix, then, the scale-free characteristic (power law of the degree distribution) of the network, is ignoring. So in order to overcome this limitation, we propose a fully Bayesian generation model, which incorporates the heterogeneity of the degree of nodes to model dynamic complex networks. Then we present a new dynamic stochastic block model for community detection and evolution tracking under a unified framework. We also propose an effective variational inference algorithm to solve the proposed model. The model is tested on the simulated datasets and the real-world datasets, and the experimental results show that the performance of it is superior to the baselines of community detection in dynamic networks.