An Agent Bilateral Multi-issue Alternate Bidding Negotiation Protocol Based on Reinforcement Learning and its Application in E-commerce

With the rapid development of multi-agent based E-commerce systems, on-line automatic negotiation protocol is often needed. But because of incomplete information agents have, the efficiency of on-line negotiation protocol is rather low. To overcome the problem, an on-line agent bilateral multi-issue alternate bidding negotiation protocol based on reinforcement learning is present. The reinforcement learning algorithm is presented to on-line learn the incomplete information of negotiation agent to enhance the efficiency of negotiation protocol. The protocol is applied to on-line multi-agent based electronic commerce. In the protocol experiment, three kinds of agents are used to compare with, which are no-learning agents (NA), static learning agents (SA) and dynamic learning agent (DA) in this paper. In static learning agent, the learning rate of Q-learning is set to 0.1 unchangeable, so itpsilas called static learning. While in dynamic learning proposed in this paper, the learning rate of Q-learning can change dynamically, so itpsilas called dynamic learning. Experiments show that the protocol present in this paper can help agents to negotiate more efficiently.