Abstract: Social networks has gained huge popularity in particular microblogs in recent years. The discovery of latent topic communities in social networks carries high value in commercial promotion, public opinion monitoring, etc. In recent years, probabilistic generative topic model (Latent Dirichlet Allocation, LDA) has been widely applied in the field of data mining. Generally, LDA can process text or digital signal data, however, without any modification, it lacks the capability to properly process the relation data between users in a social network. By modifying the original LDA model, this essay proposes a new model, Tri-LDA and applies it to dig the hidden topic communities in a social network. The experiment result shows that the topic communities found by Tri-LDA is basically consistent with the realistic topic communities that hand-labeled by the authors.

Key words: LDA, social networks, topic community