Predicting missing links in directed complex networks: A linear programming method

Abstract

Link prediction in complex networks has always been a hot topic in statistical physics, sociology and information science. Since most works focus on undirected networks, how to predict missing links in directed complex networks remains a valuable and challenging problem. Many existing methods fail to differentiate the information provided by links with different orientations, nor do they consider the unequal contributions of neighbors, leading to deficiency in prediction accuracy. In this paper, we propose a novel link prediction method in directed networks. It calculates the contributions of three types of neighbors by solving a simple linear programming problem. Empirical studies on eight real-world networks show that the proposed method performs better under two evaluation metrics in comparison with nine state-of-art benchmarks.