Link Prediction by Multiple Motifs in Directed Networks

Link prediction which can restore and predict missing links has wide applications in complex networks. In existing researches on link prediction of directed networks, most methods only consider the information of a single motif, in which the effects of multiple motifs are not included, especially the role of each node forming different motif structures. In order to solve the above problems, firstly we propose a single motif naive Bayes model beyond calculating the number of edge-dependent motifs. We also investigate a two-motif naive Bayes model and a machine learning framework based on multi-motif features to further improve the performance of link prediction. The new framework of link prediction by multiple motifs is superior to the state-of-the-art methods such as potential theory, local path and superposed random walk. Experimental results on real-life networks show the highest performance improvement is 64.3%. Finally, we use maximal information coefficients to reveal the topology correlation between different motifs, which is helpful to understand the evolutionary mechanism of directed networks.