Research On Multi-Issue Automated Negotiation Based On Agent Reinforcement Learning

With the globalization of economy,the competition of e-commerce is becoming more and more fierce.People will choose to solve the conflicts of interest.The using of agent automatic negotiation technology can replace artificial consultation time,high cost,slow reaction and other shortcomings,which helps to negotiate efficiency,and improves people's quality of life.Researching on multi-agent automated negotiation methods have obtained very many valuable research results.However,most researchers focus on the construction of the negotiation model,or using the reinforcement learning Q-learning algorithm and Baynes algorithm as a combination algorithm to solve.Known knowledge is used,but the parameters are not discussed;The combination of the reinforcement learning algorithm and the opponent classification algorithm has solved the single-issue of the consultation,but it don not slove multi-issues negotiation.So this paper is based on the existing problems.We can have the following improvements:1?Thinking reinforcement learning has some important parameters in the negotiation strategy,such as:belief knowledge,time discount rate,negotiation rounds,We propose reinforcement learning of multi-issues related negotiation algorithm.Different parameters are compared to verify that the time conviction is for decreasing function,and the discount rate is 0.9.It is the better the performance of the algorithm.2?Thinking the opponent classification algorithm and Q-learning algorithm can better adapt to the dynamic changing environment.This paper proposes an reinforcement learning based on opponent classification of multi-issues related negotiation algorithm,campared with reinforcement learning of multi-issues related negotiation algorithm.It verifies the feasibility of the algorithm.