A Markov Decision Process Approach to Analyze Discount & Reputation Trade-offs in E-commerce Systems

Abstract

E-commerce systems, e.g., Amazon, eBay and Taobao, are becoming increasingly popular. We consider eBay (or Taobao) like E-commerce systems, where a large number of sellers and buyers transact online. To reflect the trustworthiness of sellers, a reputation system is maintained. In particular, the feedback-based reputation system [4] is the most widely deployed. Sellers of such systems are initialized with a low reputation and they must obtain a sufficiently large number of positive feedbacks from buyers to earn a reputable label. For example, eBay and Taobao use three-level feedbacks, i.e., (-1 (Negative), 0 (Neutral), 1 (Positive). Each seller is initialized with a reputation score of zero. A positive (or negative) rating increases (or decreases) the reputation score by one, while a neutral rating does not change the reputation score.