A link prediction approach using semi-supervised learning in dynamic networks

Abstract

Link prediction is one of the hottest issues in social network analysis are, which aims to predict new links from a known network. This paper presents a new link prediction approach named SLiPT (Self-training based Link Prediction using Temporal features). The method introduces semi-supervised learning into link prediction task in order to use the potential information in a large number of unlinked node pairs in networks. Moreover, temporal features are used in SLiPT to improve the predictor for dynamic networks. The experimental results in two real datasets Enron and DBLP show that, the prediction accuracy of SLiPT is higher than two baseline methods SLiP (Self-training based Link Prediction) and BLiP (Basic Link Prediction) and a state-of-art link prediction method.