Unified Modeling of User Activities on Social Networking Sites

Social networking sites like Facebook and Twitter are teeming with users and the content posted by them. Several activities like friendship/followership, authoring, commenting on, liking, resharing/retweeting posts typically occur on these sites. In this paper, we make an attempt at the unified modeling of various such activities on social networking sites. We propose a novel joint latent factor model, Latent User PreferenceModel (LUPM), which combines the predictive power of multiple dyadic relations and text content using block and topic models coupled through a common latent representation for the users and posts. We further experiment with real world Twitter and Facebook datasets in order to understand the empirical significance of such unified modeling.