Uncover Product Review Patterns via Weighted Motifs

Abstract

In today’s e-commercial websites, product reviews written by genuine users are commonly seen, which play a crucial role as customer feedbacks and planting seeds to trigger much more transactions. However, motivated by profits, fake reviews crafted by spammers are inevitable to promote or demote product reputations whereas misguiding potential buyers to make bad decisions. Until recently, the problem how to distinguish whether a review is fraudulent or a reviewer is a spammer has long been studied, but the question of general review pattern mining is still open. In this paper, we model online product review systems into bipartite networks and adopt a network technique, called the weighted motif to uncover underlying reviewing patterns. Experiments on Amazon’s review dataset show that, our system is feasible and effective.