In many social networks, people interact based on their relationship network. Community detection algorithms are then useful to reveal the sub-structures of a network. Identifying these users' communities can help us assist their life-cycle. However, in certain kinds of online communities such as question-and-answer (Q&A) sites or forums, people interact based on common topics of interest, rather than an explicit relationship network. Therefore, many traditional community detection techniques do not apply directly. Discovering those topics of interest is critical to identify users' communities. Besides, users' activities on certain topics of interest are evolving with time and it is therefore very important to extract their temporal dynamics. In this paper, we first propose Topic Trees Distributions (TTD), an efficient approach for extracting topics from Q&A sites in order to detect overlapping communities. We then extend TTD to propose Temporal Topic Expertise Activity (TTEA), a graphical probabilistic model to extract both topics-based expertise and temporal information. We evaluated and compared our models with state-of-the-art approaches on a dataset extracted from the popular Q&A site StackOverflow.