基于GAI多属性依赖的协商模型

Negotiation Model Based on GAI Multi-Attribute Dependence

多属性之间的依赖关系增加协商 Agent 效用函数的复杂性,从而也增加多属性协商问题的复杂度.本文提出一种基于 GAI 多属性依赖的协商模型.该模型使用 GAI 分解将协商 Agent 的非线性效用函数表示为依赖属性子集的子效用之和.在协商过程中,协商双方采用不同的让步策略和提议策略来改变提议的内容.卖方 Agent 利用本文提出的 GAI 网合并算法将协商双方的 GAI 网合并,并利用生成的 GAI 树产生使社会福利评估值最大的提议.实验表明当买方 Agent 采用局部让步策略且卖方 Agent 采用全局让步策略时,协商双方能够在有限的协商步内达到接近 Pareto 最优的协商结局.

The interdependencies among attributes make the utility functions of agents much more complex and consequently and the multi-attribute negotiation more difficult. A negotiation model based on generalized additive independence （GAI） multi-attribute dependence is proposed. In this model, GAI decomposition is employed to represent non-linear utility functions of agents as the subutilities of subsets of interdependent attributes. During the procedure of negotiation, each negotiation agent adopts different conceding and proposing strategies to change the offers＇ content. Seller agent combines the GAI-networks of negotiation participants with the proposed combination algorithm, and produces an offer which maximizes the estimation of social welfare using the GAI tree. Experimental results show that Pareto-optimality agreements can be reached when buyer agent adopts local conceding strategy and seller agent uses global conceding strategy.