Link Prediction Based on The Similarity of Transmission Nodes of Multiple Paths in Weighted Social Networks

Most link prediction algorithms only consider local or global characteristics of the graph, so it is difficult to reach equilibrium in the accuracy and the computational complexity. And the research on link prediction in weighted networks is relatively less. A new algorithm STNMP (Similarity based on Transmission Nodes of Multiple Paths) for link prediction in weighted social networks is proposed. Firstly, the concept of the edge weight strength is introduced to measure the local similarity of neighbor node pairs. Then the definition of the path similarity contribution and the similarity of transmission nodes of multiple paths are given which are used to describe the total contribution of all these paths of 2 and 3 paces to the similarity of node pairs. The effectiveness of the algorithm is verified through experiments on many real networks using AUC and Precision as evaluation index. Furthermore, the prediction accuracy and efficiency of the algorithm are also analyzed through the comparison with those classical link prediction algorithms based on the similarity index of CN, AA, etc. The results show that for the small scale of social networks, the accuracy of STNMP algorithm is higher than those of existing algorithms. In addition, with a good generality, the algorithm is also applicable to the link prediction in unweighted social networks.