Punishing the redundant influence improves the accuracy of link prediction

The influence of traditionally composing node degree ignores the fact that effective transmission determines the similarity between endpoints in link prediction. Due to effective transmission, the degrees connected to the common neighbor and the short-path can be regarded as efficient influence, whereas the degrees connected to the noncommon neighbor and the long-path represent redundant influence because of their inefficiency, not to mention the paths disconnected to the target node. Therefore, in this paper, we propose a new method based on punishing the redundant influence (PRI) to improve the accuracy of the link prediction by emphasizing the effective influence and increasing the similarity between node pairs. From the results of extensive experiments in twelve real-world networks, PRI achieves superior performance compared to the traditional methods.