User experience in social media is characterized by rich interaction with the media content and other participants within the online community. We use statistical models to describe the patterns of music listening in online communities at different levels of model complexity. First, we adapt the LDA model to capture the users’ taste in songs and identify the corresponding clusters of media and users. Second, we define a graphical model that takes into consideration the listening sessions and captures the listening mood of users. Our session model yields clusters of media and users that capture the behavior exhibited across listening sessions, and it allows faster inference when compared to the LDA model. Our experiments with the data from an online media site (Zune Social music community) demonstrate that the session model is better in terms of the perplexity on the music genre co-occurrence compared to the LDA-based taste model that does not incorporate cross-session information and a baseline model that does not use latent clusters.