Link Prediction Based on Graph Embedding Method in Unweighted Networks

The index of link prediction based on random walk usually has the same transition probability in the process of particle transfer to its neighbor nodes, which has strong randomness and ignores the influence of the particularity of network topology on particle transition probability. In order to resolve this problem, this paper proposes a random walk with restart index based on graph embedding (GERWR). The algorithm uses graph embedding method to randomly sample network nodes and generate node representation vectors containing potential network structure information. By calculating the similarity of node vectors, it redefines a biased transition probability. We apply it to the process of random walk and explore the influence of the particularity of network topology on the transition during the particles walk. Finally, based on biased transition, the index proposed in this paper is compared with five classical similarity indexes in unweighted networks. The results show that the prediction algorithm based on graph embedding method with biased transfer has higher accuracy than other indexes.