Accurate and fast link prediction in complex networks

Link prediction is a fundamental task for analyzing complex networks which has been widely used in many domains, such as, identify spurious interactions, extract missing information, evaluate complex network evolving mechanism. There exist a variety of techniques for link prediction, ranging from node similarity-based methods to probabilistic graphical models. Node similarity-based methods have low algorithm complexity and low time consumption merit. Moreover, these methods can obtain good prediction accuracy, therefore node similarity-based method have become the mainstream technique. However, most of similarity-based link prediction methods only take into account the role of each common neighbor equally to the connection probability of two nodes. In addition, these methods only take into account the contribution of each 2 hops common neighbor. In fact, 3 hops common neighbor also give valuable contributions to the connection likelihood. In this paper, we propose a model for link prediction, which is based on 2 and 3 hops common neighbors. In our model, each 2 or 3 hops common neighbor plays a different role to the node connection probability according to their degrees. Extensive experiments were conducted on six real-world networks. Compared with the representative node similarity-based methods, our proposed model can provide more accurate predictions.