An Improved Link Prediction Algorithm Based on Comprehensive Consideration of Joint Influence of Adjacent Nodes for Random Walk with Restart

Abstract

In the basic random walk link prediction method, the probability of a walking particle when selecting a neighbor node for a walk is determined only by the degree of the current node, and it is fixed and uniform, without considering the impact of degree of the neighboring nodes on the transition probability. In view of this, a link prediction algorithm is proposed in which the degrees of the current node and its neighbor nodes jointly determine the transition probability. First, using the transition probability model of Metropolis-Hasting Random Walk (MHRW) algorithm to redefine the transition probability of the walking particles between the neighbor nodes, then combining Random Walk with Restart (RWR) similarity index to propose the Metropolis-Hasting Random Walk with Restart (MHRWR) algorithm in this paper for link prediction. The link prediction comparison experiments been performed on 6 different scale real network datasets. Compared with the benchmark algorithm, the MHRWR algorithm not only improved the AUC index, but also improved the Precision and Ranking score; compared with the RWR algorithm, the AUC value has increased by an average of 2.10%, and the highest is 5.34%. Experimental results show that the MHRWR algorithm of our proposed leads to superior accuracy in link prediction.