Abstract: It is very important to discover closely-connected interactive communities, as well as mining their evolution patterns, by analyzing the content information, to benefit network marketing, content recommendation and other online applications. To the best of our knowledge, most of the hybrid models based on the content and link analysis failed to integrate the timestamp information into a unified model, and thus failed to use the temporal information as well. In this paper, we propose a new community discovery model, COT(community over time), based on the Bayesian graphical model, which can integrate textual content, topology structure and temporal information. Experimental evaluation on a real dataset from the SinaWeibo was performed, and the result shows that our model has obvious effect for online social networks and obtains better interpretability.