A Reputation-Oriented Reinforcement Learning Approach for Agents in Electronic Marketplaces

Abstract

The problem of how to design personal, intelligent agents for e-commerce applications is a subject of increasing interest from both the academic and industrial research communities. In our research, we consider the agent environment as an open marketplace which is populated with economic agents (buyers and sellers), freely entering or leaving the market. The problem we are addressing is how best to model the electronic marketplace, and what kinds of learning strategies should be provided, in order to improve the performance of buyers and sellers in electronic exchanges. Our strategy is to introduce a reputation-oriented reinforcement learning algorithm for buyers and sellers. We take into account the fact that multiple sellers may offer the same good with different qualities. In our approach, buyers learn to maximize their expected value of goods and to avoid the risk of purchasing low quality goods by dynamically maintaining sets of reputable sellers. Sellers learn to maximize their expected profits by adjusting product prices and by optionally altering the quality of their goods. In our buying algorithm, a buyer b uses an expected value function f b , where f b (g, p, s) represents buyer b’s expected value of buying good g at price p from seller s. Buyer b maintains reputation ratings for sellers, and chooses among its set of reputable sellers S b