With the rapid popularity of Web 2.0 techniques, community question answering (CQA) services such as Yahoo! Answers have recently become a novel platform for knowledge sharing. Given their interactivity and openness, these services can well serve web users’ personalized information needs. A typical CQA system consists of three elementary actors, including users, questions and answers. Exploring the relationships among these actors and analyzing users’ behavior patterns can improve service quality and enhance users’ loyalty. The contents of questions or answers submitted by CQA users reflect their topical interests. Users’ participations in a certain topic can accurately reveal the development of the related social events. Thus, exploiting the topic structures in CQA, especially the hot topics that are highly concerned by public, is beneficial to detect public opinions, leading to better informed decisions and more effective policy implication. Users who show common interests in a certain topic tend to form a densely connected user community by interacting regularly. Detecting user communities will help analyze user behaviors from a macro point and find authorized users in the community topics, among others. Since users in the same community share similar interests, they are more likely to establish friend relationships. On the basis of detected user communities, recommending latent friends can enhance the information exchange among users, hence improving knowledge sharing and spreading. However, the current researches on CQA mainly focuse on the micro-level, such as evaluating question and answer qualities, retrieving similar questions or predicting user satisfactions. The topic structures, user communities and friendship relationships among others have received little attention. Based on web mining techniques, this thesis studies CQA from these three major standpoints, i.e. topics, user communities and latent friends. Specifically, the main research focuses are summarized as follows. 1) We present a framework for hot topic detection and trend analysis, which extracts the hot topics during certain time period and tracks the variation of given event or topic. It will help identify the information aggregation which attracts broad and continuous concerns and reveals the public opinions. By exploring the temporal characteristics of keywords, it can accurately identify hot terms. Topic clustering can help exploit the topic focuses of certain topic. Consi...
修改评论