A Multi-Agent Based Electronic Commerce System Based on Classified Learning Mechanism
In multi-agent based electronic commerce systems, on-line negotiation is often needed, but because of incomplete and asymmetric information agents have, an agent may propose infinite proposition and counter proposition, the deal is also not reached, so the efficiency of agent negotiation is rather low. In order to decrease the incomplete and asymmetric information between agents in multi-agent based electronic commerce, accelerate the process of agent negotiation, and enhance the efficiency of agent negotiation, a multi-agent based electronic commerce system based on classified learning mechanism is presented. The learning mechanism is presented to learn the incomplete and asymmetric information, especially learn the issue weight of the agents opponent. The learning principle is which issues negotiation concession is more, and then the issues weight is lower. From the point of view of statistical negotiation concession, the issue weight of agents opponent can be learned approximatively. In the experiment, the bilateral multi-issue negotiation protocol is used and two kinds of negotiation agents are used lo compare, one is the agent with no learning (NLA), the other is the agent with classified learning mechanism(CLA). After the two agents reach the satisfying result, the NLA uses 1038 runs of negotiation, while the CLA only uses 303 runs of negotiation. The experimental results show that the agent with learning mechanism can negotiate more efficiently than that of no learning agent
Electronic commerce Multi-agent systems Incomplete and asymmetric information Learning mechanism Negotiation
Jing Bo Li Jian
School of Computer Beijing University of Posts and Telecommunications Beijing, China Dept of Compute School of Computer Beijing University of Posts and Telecommunications Beijing, China
国际会议
成都
英文
1171-1175
2010-12-17（万方平台首次上网日期，不代表论文的发表时间）