A similarity index algorithm for link prediction

Abstract

Link prediction in networks is that using the existing known network structure or node infor­mation to predict the possibility between the two nodes which haven't connected to each other. It's important to learn about the evolution mechanism of network and the interaction relationship of nodes. The link possibility between nodes is closely related to the similarity. The method which is based on the node attributes and local information has the simple and direct calculation and better effect of prediction. So it is more suitable for the large-scale network applications. But it only considers the degree of final nodes or neighbor nodes and the number of neighbor nodes. Does not take into account that each neighbor nodes has the different effect for the different final nodes. The paper through experiments to analysis and compare different similarity contribution of neighbor nodes and end points. And further verified the weak-link effect in networks. Also we proposed a new common neighbor measurement algorithm, through distinguish the influence of each common neighbor for the different end nodes so that the prediction accuracy has been further improved.