Complements or confounders? A study of effects of target and non-target features on online fraudulent reviewer detection

Abstract

The online review fraud process is an event in which a reviewer posts fraudulent reviews on an e-commerce platform with respect to a review target, such as a commodity or service. Extant studies on detecting fraudulent reviewers often rely on reviewer behavioral patterns and the textual content of reviews while ignoring the targets being reviewed. Based on the Goals-Plans-Action theory, we examine the relative importance of target features for fraudulent reviewer detection in comparison to that of non-target features. Target features refer to the features derived from the innate information related to the reviewed products or services while non-target features refer to the features derived from the acquired information related to the reviewed products or services. In this study, we analyze a sample from the Yelp.com dataset of restaurant reviews to help better understand the importance of target features and non-target features. The results suggest that using the combination of target features with non-target features can improve the performance of online fraudulent reviewer detection in comparison with using non-target features alone. Moreover, using the combination of target features with non-target features will further improve the performance of online fraudulent reviewer detection when we consider the non-target features as conditioned on the reviewed target features rather than treating them as independent of each other.