Analysis of User Behavior against Opinion Spammer: Following and Correcting Effect

Abstract

Opinion Spam is still a widespread problem in online review platforms. Opinion spam is hard to detect, because of spammers sophisticated strategy to avoid detection system. In contrast with other kind of spams, context is not a powerful feature to detect opinion spam. This is the most challenging point of detecting opinion spam. In this thesis, we analyzed opinion spams effect from the perspective of truthful users reactions. We found out a timing of spammers attack and also showed an activity of users is increased after the attack. Normally users agreed with previous reviewers opinion and we observed the phenomenon became more evident when spammers attacked a product. And we found out that there are both following action and correcting action of truthful users who are affected by spammers. After the attack, some of truthful users are hooked to spammers and follow them, whereas, some others try to remedy contaminated online society. We used Yelp dataset to analyze temporal dynamics around spammers and revealed these significant signals with empirical and statistical probability. This is the first research analyzed truthful users action responding to opinion spammers. We also detected spammers attack period using our new observations. As a result, we identified spammers attack strategy, effect, discovered truthful users action patterns (following and correcting effect) responding to spammers, and effectively detected spammers attack period.