An Efficient Link Prediction Method using Community Structures

Abstract

The problem of link prediction/recommendation requires to evaluate the scores of $O(n^{2})$ node pairs. While this exhaustive search could be computationally very expensive, it might also produces many zero links scores. In this paper, we propose a simple, efficient, and scalable link prediction method based on network communities. Given a complex network with community structures, the global link prediction problem is divided into several sub-problems. Each sub-problem is respon-sible for performing link prediction inside each community. The outputs of the sub-problems are combined to the final high-scored links. The results on several complex networks show the efficiency of the proposed method without sacrificing its prediction accuracy.