Research On Multi-Issue Automated Negotiation Method Based On Agent

With the rapid development of Internet, electronic commerce arises at the historic moment. It not only improves the efficiency of people's live and saves a lot of cost,but also improves the production capacity of the society. Negotiation is an important means to realize online bargaining of e-commerce, and also an important goal that people design intelligent Agent. Now there has been a lot of methods researching on Agent automated negotiation. But the developed methods just consider the improvement of the single negotiation performance such as the equality of negotiation, the utility of negotiation, the negotiation time and so on,which would always affect the other performance accordingly under improving some negotiation performance; or always ignore the correlation between issues; Or the negotiation model is much idealistic, which does not conform to the negotiation in real life. Thus,according to the advantages and disadvantages of the existing research on negotiation, this article is to improve the negotiation method and to optimize the negotiation model of Agent on the negotiation framework, negotiation strategies, the adaptive learning ability of Agent and so on.The main work includes the following four parts:(1) Considering the equality of negotiation, this paper puts forward a bilateral multi-issue parallel negotiation framework based on mediator Agent, mainly including three Agent: the buyer Agent, the seller Agent and the mediator Agent. The buyer Agent and seller Agent submit their proposals to the mediator Agent at the same time, and the mediator Agent need to judge whether there is a trading opportunity and determine the final negotiation agreement.(2) To optimize the negotiation strategy of both Agents, using reinforcement learning algorithm (Q-learning algorithm) to generate the optimal negotiation strategy dynamically,improving the negotiation performance. And carries on a contrast experiment with the model'A simultaneous multi-issue negotiation through autonomous Agents', the method is proven to really improve the overall performance of the negotiation.(3) To optimize the traditional reinforcement learning negotiation strategy, introducing the parameter expected reduction rate to restore the original expected utility, which avoid the negotiation Agents making too much concession in the beginning. And has carried on a contrast experiment, the experimental results show that this method is not only effective, but also efficient.(4) To optimize the ability of adaptive learning and mediation of the mediator Agent,introducing a benchmark concessions utility function .The mediator Agent adjusts the buyer and the seller Agent's negotiation strategy through the benchmark function to coordinate the negotiation between both Agents. Finally the method is verified by experiments that did promote the consultation process, and optimize the overall performance of the negotiation.