Link Prediction Based on Generalized Cluster Information

Abstract

Understanding of which new interactions among data objects are likely to occur in the future is crucial for a deeper understanding of network dynamics and evolution. This question is largely unexplored except a local neighborhood perspective, partly owing to the difficulty in finding major factors which heavily affect the link prediction problem. In this paper, we propose LPCSP, a novel link prediction method which exploits the generalized cluster information containing cluster relations and cluster evolution information. Experiments show that our proposed LPCSP is accurate, scalable, and useful for link prediction on real world graphs.