基于Bayesian学习的适应性优化协商模型

在复杂的自动协商环境中，设计能够处理不完全信息和动态情形的协商agent有效学习机制正成为具有挑战性的议题.提出了一种基于Bayesian学习的时间依赖的双边多议题协商优化模型(BLMSEAN).通过只观察对手的历史报价，将Bayesian学习和基于混合策略的演化算法相结合，所提模型使得协商agent能够对于对手协商参数的概率分布有更精确的估计(如期限、保留报价和议题权重等)，能够适应性地调整让步策略使协商双方都受益，提高了协商的成功率和效用.通过实验可以显示所提的模型学习对手私有信息和适应性调整让步策略的有效性.

An Adaptive and Optimized Negotiation Model Based on Bayesian Learning

In complex automated negotiations, a challenging issue is how to design effective learning mechanisms of agents that can deal with incomplete information, in which the agents do not know the opponent’s private information (i.e., the deadline, reservation offer, issue weight) and such information may be not unchanged. We present a time dependent, bilateral multi-issue optimized negotiation model by combining Bayesian learning with evolutionary algorithm based on mixed strategy (BLMSEAN). The proposed model defines reservation units, reservation points, and the each probability of reservation point which can be on behalf of the likelihood of the reservation point located in the unit. A regression analysis compares the correlation between estimated offers and historical offers, and Bayesian rule updates the probabilities and the weights of issues utilizing the historical offers only. The evolution algorithm with mixed mutation strategy enables the estimation to approximate more accurately opponent’s negotiation parameters and to adjust adaptively concession strategy to benefit two partners to improve the joint utility and success rate of negotiation agreement. By being evaluated empirically, this model shows its effectiveness for the agent to learn the possible range of its opponent’s private information and alter its concession strategy adaptively.