Improved Link Prediction in Social Networks using Label Propagation

Abstract

Social communications can effectively assist social networks in order to keep their active users and improve services of social networks. Therefore, they should have good prediction of social communications between their users. In this way, link prediction issue in social networks has interested many researchers and increasing prediction accuracy in social networks in under focus. There are three approaches for solving link prediction problem: similarity-based approach, an approach based on maximum likehood and probabilistic model-based approach. In this paper, similarity-based approach is utilized. Also, label propagation concept is used for link prediction and in order to raise accuracy in link prediction algorithm by using label propagation, the distance of the shortest path is applied for label propagation in network. Presented method, with AUC metric is tested on football data set, Email network, science network and power network. Total results of this test show that proposed method is improved basic method about 17.133%.