基于对手态度学习的协商决策模型  

Negotiation decision model based on learning of opponent's attitudes

作　　者：程昱[1] 高济[1] 古华茂[2] 傅朝阳[1] 

机构地区：[1]浙江大学人工智能研究所,浙江杭州310027 [2]浙江工商大学信息与电子工程学院,浙江杭州310018

出　　处：《浙江大学学报（工学版）》2008年第10期1676-1680,共5页Journal of Zhejiang University：Engineering Science

基　　金：国家“973”重点基础研究发展规划资助项目(2003CB3l7000);浙江省自然科学基金资助项目(Y106369);浙江省科技厅支撑和引导计划资助项目(2007C21042)

摘　　要：为了解决基于多Agent应用环境下的双边自动协商问题,提出了一种用基于支持向量机(SVM)算法的对手协商态度学习方法.在该方法中,协商的过程被看作一个建议序列,把建议序列映射到新的特征空间,形成了多个协商轨迹(每个协商项有一个协商轨迹).通过支持向量机的方法来学习协商轨迹,得到协商对手在每个协商项的态度.然后利用学习得到的对手协商态度,构造了一个协商的决策模型.此模型能同时基于对手的态度和自身的偏好来做出协商决策.另外,模型中的模型函数在满足一定约束的条件下,可以保证协商决策的收敛性和单调性.实验结果表明,该模型能有效较少协商的时间,增加协商双方的效用总和.A support vector machine （SVM） based method was proposed to learn opponent＇s attitudes to solve the problem of bilateral automated negotiation in agent-mediated application. The procedure of negotiation was viewed as a proposals＇ sequence which can be transformed to multiple negotiation tracks-- one negotiation track for each negotiation issue-- by mapping them to a new feature space. Then the opponent＇s attitude of each issue can be got by learning the negotiation tracks. A negotiation decision making model was constructed by utilizing the opponent＇s attitude. The model can make effective trade-offs between opponent＇s attitudes and self＇s preferences. If the model＇s function constraints to some condition, the convergence and monotonic of model can be promised. Experimental results show that the model can decrease negotiation time and increase the total utility of negotiation participants.

关 键 词：多AGENT系统 自动协商 支持向量机 协商态度 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]