Web User Trust Relationship Prediction based on Evidence Theory

Abstract

Mining web user trust relationship is important in web information credibility analysis. Motivated by the imprecise nature of trustiness, we propose a novel web user trust prediction method based on evidence theory, which uses user ratings to infer trust relationships between users, where each rating score is treated as an evidence. The method first generates basic probability assignments with training sample set using classical statistics; then, integrates conflict resolution rules and D-S evidence combination rules to obtain global BPA; finally, identifies user trust relationships based on global BPA. Our experiments conducted on the Extended Epinions dataset show that the precision, recall and F-measure of the proposed approach reaches 96.78%, 99.44% and 98.10% respectively, with 0.06%, 13.41% and 9.09% improvement over current best practices, demonstrating the advantages and effectiveness of using evidence theory to identify user trust relationship in social networks.