Attractive density: A new node similarity index of link prediction in complex networks

Abstract

Link prediction plays an important role in data mining, which aims at estimating the probability of the connection between two unlinked nodes according to the information of network structure. Many link prediction methods have been proposed so far, while most of them only consider the node similarity based on individual information of common neighbors. In the perspective of interactions between common neighbors, we present a new node similarity measurement—inner attractive density of the cluster formed by common neighbors. The proposed index not only applies the effect of individual node in common neighbors set, but also considers the interactions among common neighbors. Experimental results on synthetic and real networks show that compared with the typical prediction algorithms, the proposed method enjoys impressive efficiency and effectiveness, and it improves the accuracy of prediction while maintaining low time complexity.