Abstract: It is a critical issue to detect dynamic communities and track their evolution process in online social networks, which can help the controller understand the latent topology, discover anomaly events, predict its evolution trend and control the networks. Firstly, the current flaws of dynamic community detection and its evolution were analyzed. And then a novel approach of dynamic evolution of communities was proposed, including community extract in each time snapshot based on a static community mining algorithm, the calculation of evolution influence between the neighboring snapshots in the community, and generating the evolution process of community structure among continuous snapshots. Finally, tests were carried out based on the large-scale data-sets (e.g. Micro-blog, Gnutella) to validate the approach. The results show the high effectiveness of the approach in community evolution analyzing. In addition, the experiments also analyze the frequency, at which the social network nodes appear and disappear, will affect the community stability and the evolution of the structure.