A Unified Topic-Style Model for Online Discussions

Forums have become major places for online communications for many years, where people often share and express opinions. We observe that, when editing posts, while some people seriously state their opinions, there are also many people playing jokes and writing meaningless posts on the discussed topics. We design a unified probabilistic graphical model to capture both topic-driven words and styledriven words. The model can help us separate serious and unserious posts/users and identify slang words. An extensive set of experiments demonstrates the effectiveness of our model.