A RECURSIVE SEQUENTIAL BAYESIAN APPROACH FOR THE LINK PREDICTION PROBLEM

摘要

Link prediction problems are being Increasingly adopted to detect the potential links In various domains. One challenging problem Is to Improve the accuracy continually. Based on the Idea of sequential Bayeslan updating method, the authors have revealed a novel approach which finds a posterior based on the observed data, assesses the state of a graph and uses this posterior as a prior distribution for the next stage. The proposed approach Incorporates more topological structure Information and node attributes data with increasing Iterations. Experimental results with real-world covert networks have shown the proposed method performs better In terms of evaluation metrics in comparison with other methods. Numerical experiments are conducted on terrorism networks. The proposed approach can provide the decision-makers with effective auxiliary information and proves to be a perspective tool in link prediction problems.