Link Prediction of Directed Network Based on Node Importance

Directed link prediction aims to predict the emergence of future relations and their orientations from the structure information of current network. A lot of methods have been proposed for directed link prediction, however, most of them only consider the shared information between node pairs when calculating the node similarity, and ignore the difference in the contribution of the importance of the node itself to the formation of the link. For this reason, in this paper, we propose a method based on the importance of the node for directed link prediction. It not only considers the shared information but also the nodes' importance. The method utilizes the PageRank algorithm to calculate the PR value for every node and corrects it as the importance of the node by the network size, and then integrates the importance of the node into the adjacent matrix of the network by defining the reciprocal link weight. Then we use the improved Bifan predictor to calculate the similarity of node pairs for link prediction. To validate the introduction of the nodes' importance, we further extend our method to classic predictors and experiments results on eight real directed networks demonstrate that our method improves the prediction accuracy.