Detect Online Review Spammers Based on Comprehensive Trustiness Propagation Model

Abstract

Review spammers detection is an important task in social media sentiment analysis. Previous works employ reviewer behaviors such as text similarities, duplications and rating patterns to indentify suspicious spammers. However, there are still other kinds of abnormal spamming activities which could not be detected by the available techniques. This paper proposes a review spammer detection approach combining both TrustRank and Anti-TrustRank propagation algorithm to identify review spammers. Firstly, a twolayer heterogeneous review relation graph is constructed to capture the relationships among reviewers and products. Secondly, a TrustRank based propagation model and an Anti-TrustRank based propagation model are established to calculate the reviewers’ trustiness value and the reviewers’ anti-trustiness value respectively. Finally, review spammers are detected according to the comprehensive trustiness value which combines both reviewers’ trustiness value and anti-trustiness value. Experimental results show that according to two datasets, our presented method significantly outperforms the existing baselines, and is able to find more abnormal spamming activities.