论文部分内容阅读
文中提出了一种预测协商中Agent行为的学习机制,该机制的基础是仅使用协商交往中对方的历史响应进行非线性回归分析。自动协商中对方Agent的行为由其决策函数表示的策略决定。先通过一系列的模拟得到对方在采用各种策略和参数配置的响应,然后总结提取了估计对方策略的启发性知识,最后把此知识应用到实验性的在线协商中进行测试。结果表明使用这些知识能够取得比现有决策函数策略更好的结果。该学习机制可以在线使用,也不需要有关于对方的过去知识,在双方不了解或很少了解的开放式系统中尤为有效。
In this paper, we propose a learning mechanism that predicts Agent behavior in negotiation. The mechanism is based on the nonlinear regression analysis of each other’s historical responses using only the negotiation interactions. The behavior of the partner agent in autonegotiation is determined by the strategy indicated by its decision function. First, through a series of simulations, we get the responses of each other using various strategies and parameters, and then summarize and extract the heuristic knowledge of estimating each other’s strategies. Finally, we apply this knowledge to the experimental online negotiation. The results show that using this knowledge can achieve better results than the existing decision function strategy. This learning mechanism is available online and does not require past knowledge of each other and is especially effective in open systems that either do not understand or know little about.