Trust-Aware Review Spam Detection

Abstract

Online review systems play an important role in affecting consumers' behaviors and decision making, attracting many spammers to insert fake reviews to manipulate review content and ratings. To increase utility and improve user experience, some online review systems allow users to form social relationships between each other and encourage their interactions. In this paper, we aim at providing an efficient and effective method to identify review spammers by incorporating social relations based on two assumptions that people are more likely to consider reviews from those connected with them as trustworthy, and review spammers are less likely to maintain a large relationship network with normal users. The contributions of this paper are two-fold: (1) We elaborate how social relationships can be incorporated into review rating prediction and propose a trust-based rating prediction model using proximity as trust weight, and (2) We design a trust-aware detection model based on rating variance which iteratively calculates user-specific overall trustworthiness scores as the indicator for spamicity. Experiments on the dataset collected from Yelp.com show that the proposed trust-based prediction achieves a higher accuracy than standard CF method, and there exists a strong correlation between social relationships and the overall trustworthiness scores.