Roles of degree, H-index and coreness in link prediction of complex networks

Abstract

Link prediction based on topological similarity attracts more and more interests. Traditionally, researchers almost focus on utility of the paths between two unlinked endpoints, but pay little attention to the influence of the endpoints with only degree considered. Through profound investigations, we find, besides of degree, H-index and coreness also can play important roles in link prediction as the influence of endpoint especially in models based on representative SRW which is for the first time introduce influence into link prediction. In this paper, we mainly research degree, H-index and coreness in SRW-based models to explore their roles in accurate link prediction. Extensive experiments on twelve real benchmark datasets suggest that in most cases H-index serves as a better tradeoff in accurate link prediction than either degree or coreness.