Extension of neighbor-based link prediction methods for directed, weighted and temporal social networks

Abstract

Link prediction is one of the most interesting tasks in social network analysis. It has received considerable attention as evident by the number of studies described in the literature. Recently, heterogeneous, temporal or directed based network models have attracted considerable attention to deal with effectively real complex networks in terms of link prediction. Most of the link prediction measures in the literature don't consider the role of link direction. In this study, we introduce a directional link prediction measure by extending neighbor based measures as directional pattern based to take into account the role of link direction in directed networks. The introduced measure also considers weight and time information of links, which are effective to improve accuracy of link prediction. In experiments, the introduced measure is compared to nine well-known link prediction measures in the literature by using supervised learning algorithms. Experimental results demonstrate that the proposed approach improves remarkably the accuracy of link prediction. This is mainly due to using structural information of networks effectively without requiring more information and computational time. 2018. (C) 2018 Elsevier Inc. All rights reserved.