Time evolution is one important feature of communities in network science. It is related with capturing critical events, characterizing community members, and predicting behaviours of communities in networks with time varying. However, most of existing community detection techniques are proposed for static networks. Here, we present a new framework to uncover community structure for each temporal graph over time. In consideration of regularizing time-dependent communities, the high temporal variations will be prevented and the gained results on community evolution become more reasonable. Having applied it on synthetic networks, the experimental results offer new views in dynamic networks