Predicting Links and Inferring Attributes using a Social-Attribute Network (SAN)

The effects of social influence and network autocorrelation suggest that both network structure and node attribute information should inform the tasks of link prediction and node attribute inference. However, the algorithmic question of how to efficiently incorporate these two sources of information remains largely unanswered. We propose a Social-Attribute Network (SAN) model that gracefully integrates node attributes with network structure to predict network links and infer node attributes. We adapt leading supervised and unsupervised link prediction algorithms to the SAN model and demonstrate performance improvement for each algorithm. We then show that link prediction accuracy is further improved by first inferring missing attributes. We evaluate these algorithms on a novel Google+ network dataset and achieve state-of-the-art link prediction and attribute inference performance.