Social networks have attracted particular attention in recent years, largely because of their critical role in various applications.With the advent of online social networking, the automatic discovering communities is vital for understanding the cooperation and interaction patterns of users in these social networks.However,most existing social network analysis approaches focus on investigating static topics or community structure without considering their evolutions.In this paper we study how to discover the co-evolution of topics and communities over time in dynamic social networks.We present a dynamic topic model-based approach that automatically captures the dynamic features of communities and topics evolution.Our model can be viewed as an extension of the LDA(Latent Dirichlet Allocation) model with the key addition that it can not only detect communities and topics simultaneously but also tracking how the community structure and topic changes over time.Instead of modeling communities and topics in statistical manner, the proposed model can simulate the users interests drifting at different time epochs by taking into consideration the temporal information implied in the data, and observe how the community structure changes over time with the evolution of topics.Experiments on real-world data set have proved the ability of this model in discovering well-connected and topically meaningful communities and the co-evolution pattern of topics and communities.