基于增强学习协商策略的研究及优化  

Research on reinforcement learning negotiation strategy and its optimization

作　　者：孙天昊[1] 邓俊昆[1] 陈飞[1] 朱庆生[1] 

机构地区：[1]重庆大学计算机学院,重庆400030

出　　处：《计算机工程与应用》2012年第23期44-46,51,共4页Computer Engineering and Applications

基　　金：中央高校基本科研业务费科研专项项目(No.CDJZR10180014)

摘　　要：增强学习在电子商务中可以帮助Agent选择最优行动,并达成目标。在传统增强学习协商策略中,Agent一开始便进行大幅度的妥协,这是不合理的,与现实不符,降低了Agent的期望。通过期望还原率来还原Agent的真实期望,对协商策略进行优化;讨论了期望还原率的取值对协商过程的影响;通过实验验证了优化的协商策略在保证协商效率的同时,提高了协商解的质量。Reinforcement learning can help negotiation agent to select its best actions and reach its final goal.Agent of traditional reinforcement learning negotiation strategy significantly compromises at the beginning of negotiation,which is irrational,loses touch with reality,and greatly reduces expectation of Agent.Expectation restoration rate is used to restore the true expectations of agent to optimize the negotiation strategy;the impact of value of expectation restoration rate on negotiation process is discussed;experimental results show that optimized negotiation strategy improves the quality of the negotiation result,while ensuring negotiation efficiency.

关 键 词：协商策略 增强学习 期望还原率 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]