Abstract
With facilities provided by Internet, people who connect to it could benefit from rapid information exchange without geographical limitation. And many applications are emerging to provide Internet with more possibilities and visions. A special topic about how intelligent agents represent customers to automatically accomplish tasks has recently attracted a lot of attentions both in the field of artificial intelligence research, and electronic commerce.The topic, which we are interested in, is how to design a trading agent that automatically trades in electronic markets and reveals preferences of customers. We use a synchronous double auction as a simulating test-bed, and design a reinforcement learning agent to accomplish the task. And we conduct a series of experiments with different settings to investigate the performance of our learning agents. As the evidence provided by our experimental results, we show that agents could benefit from learning when the opponent that they are against with has homogeneous behavior and unified strategy, but fail to get further benefits when the market is full of heterogeneous agents with diverse behaviors.