It is well known that users' behaviors (actions) in a social network are influenced by various factors such as personal interests, social influence, and global trends. However, few publications systematically study how social actions evolve in a dynamic social network and to what extent different factors affect the user actions.

In this paper, we propose a Noise Tolerant Time-varying Factor Graph Model (NTT-FGM) for modeling and predicting social actions. NTT-FGM simultaneously models social network structure, user attributes and user action history for better prediction of the users' future actions. More specifically, a user's action at time t is generated by her latent state at t, which is influenced by her attributes, her own latent state at time t - 1 and her neighbors' states at time t and t - 1. Based on this intuition, we formalize the social action tracking problem using the NTT-FGM model; then present an efficient algorithm to learn the model, by combining the ideas from both continuous linear system and Markov random field.

Finally, we present a case study of our model on predicting future social actions. We validate the model on three different types of real-world data sets. Qualitatively, our model can uncover some interesting patterns of the social dynamics. Quantitatively, experimental results show that the proposed method outperforms several baseline methods for action prediction.