A general spammer indicator of rating systems uncovering rating preferences and bias

Abstract

With the rapid development of e-commerce recently, massive spammers purposely distort the ranking of goods, which affects the market order and the fair competition of businesses seriously. Therefore, identifying such spammers is significant to rational decision making of customers. However, it is difficult to discriminate between normal users and malicious spammers on extremely large rating networks. In this paper, we propose a common indicator based on the historic rating records from spammers and normal users, which is widely applied to many existing methods. It is inspired by the idea that normal users have their preferences and rating bias shown in rating ladder, while spammers do not have such rating ladders in practice. Such an indicator is complement with other existing methods including Deviation-based Ranking (DR), Iterative Group-based Ranking (IGR) and Iterative Balance Ranking (IBR). Experimental study on three real rating networks shows that this indicator can significantly improve the accuracy of DR, IGR and IBR. To deal with malicious spammers, DR, IGR and IBR are improved by at least 9.38%, 2.90% and 2.53%, respectively. To deal with random spammers, DR, IGR and IBR are improved by at least by 5.52%, 17.12% and 32.24%, respectively.