Building Robust Reputation Systems in the E-commerce Environment

In the environment of e-commerce, building a robust Reputation System is a task of not only constructing an effective reputation model in terms of manipulation resistance, but also designing efficient mechanisms to detect customers and vendors whose behavior arouses suspicion. Given context of reputation, we can extract characters or assumptions from the target system. Based on these characters, an advanced reputation model, R-Rep, is proposed to resist manipulative behavior. Clustering algorithm k-means is used to identify suspicious vendors and customers. Results indicate that, R-Rep outperforms two existing models, the reputation model employed by Taobao (the largest e-commerce site in China) and a Bayesian System. By using R-Rep, negative influence imposed by suspects on Reputation Systems is largely restricted over time. Meanwhile, via comparing between statistics of the whole population and suspicious subpopulation, we explore the characteristics of suspicious customers and suspicious vendors intensively, and discover some interesting patterns and phenomena.