Exploring Heterogeneous Product Networks for Discovering Collective Marketing Hyping Behavior

Online spam comments often misguide users during online shopping. Existing online spam detection methods rely on semantic clues, behavioral footprints, and relational connections between users in review systems. Although these methods can successfully identify spam activities, evolving fraud strategies can successfully escape from the detection rules by purchasing positive comments from massive random users, i.e., user Cloud. In this paper, we study a new problem, Collective Marketing Hyping detection, for spam comments detection generated from the user Cloud. It is defined as detecting a group of marketing hyping products with untrustful marketing promotion behaviour. We propose a new learning model that uses heterogenous product networks extracted from product review systems. Our model aims to mining a group of hyping activities, which differs from existing models that only detect a single product with hyping activities. We show the existence of the Collective Marketing Hyping behavior in real-life networks. Experimental results demonstrate that the product information network can effectively detect fraud intentional product promotions.