User Topic Discussion Modeling And Its Applications In Online Forums

With the flourish of Web 2.0 applications, we have witnessed a great deal of online social medias, e.g., online forums, weblogs, question-answering communities, have emerged and thrived to become popular among massive users. Among these prevalent social medias, online forums are characterized as a unique type of platforms for news publication, hobby sharing and knowledge seeking. Popular forums usually have millions of users, with a huge amount of discussion threads updating every day.In the form of topic discussions, users interact with each other to exchange information and share knowledge in online forums. Modeling the evolution of topic discussion reveals how information propagates on Internet. It can thus help understand sociological phenomena and improve the performance of applications such as opinion mining and recommendation systems. In this paper, we argue that a user's participation in topic discussion is motivated by both her friends and her own preferences. Inspired by the theory that information flows through underlying influential network, we propose dynamic topic discussion models by mining influential relationships of users as well as individual preferences. Reply relations of users are exploited to construct the fundamental influential network. Furthermore, we consider different user interaction patterns associated with different latent topics. Based on the observation that most threads have similarity relations with each other, we propose a novel topic modeling method called Locally Discriminative Topic Modeling for thread content semantics mining and latent topic-level user discussion modeling. Additionally, the time lapse factor are also considered in order to model the changes of user interests. Based on the link analysis, we propose a novel measure called ParticipationRank to rank users according to how important they are in the social network and to what extent they prefer to participate in the discussion of a certain discussion topic. This ranking can be directly used to predict user participation in future discussions.The experiments have shown our model can simulate the evolution of topic discussion well and predict the tendency of user participation accurately. The modeling result can be applied to personalized recommendation system. We have observed the participation behaviors of the users are not only related to individual preference, but also related to interaction relationships of users. By using topic modeling method to analyze the topic-level user participation, we have observed user discussion is latent topic-specific. It also has been shown that user interests in discussion topic may change over time.