Improving the performance of link prediction by adaptively exploiting multiple structural features of networks

So far, many network-structure-based link prediction methods have been proposed. However, these traditional methods were proposed by highlighting one or two structural features of networks, and then use the methods to implement link prediction in different networks. In many cases, the performance is not ideal since each network has its unique underlying structural features. In this article, by analyzing different real networks, we find that the structural features of different networks are remarkably different. In particular, even in the same networks, their inner structural features are utterly different. Inspired by these facts, an \emph{adaptive} link prediction method is proposed to incorporate multiple structural features from the perspective of combination optimization. In the model, the weight of each structural feature is \emph{adaptively } determined by logistic regression but not be artificially given in advance. According to our experimental results, we find that the logistic regression based link prediction outperforms other typical similarity indices.