On learning mixed community-specific similarity metrics for cold-start link prediction

We study the cold-start link prediction problem where edges between vertices is unavailable by learning vertex-based similarity metrics. Existing metric learning methods for link prediction fail to consider communities which can be observed in many real-world social networks. Because dierent communities usually exhibit dierent intra-community homogeneities, learning a global similarity metric is not appropriate. In this paper, we thus propose to learn community-specific similarity metrics via joint community detection. Experiments on three real-world networks show that the intra-community homogeneities can be well preserved, and the mixed community-specific metrics perform better than a global similarity metric in terms of prediction accuracy.