基于马尔可夫链蒙特卡罗方法的双边多议题协商模型

摘要

为了提高自动化双边多议题协商的成效,提出了建立贝叶斯后验模型,以协商历史数据为训练样本,学习对手的协商偏好,依据对手偏好制定双赢的协商反建议,进而提高协商成效。假设空间是复杂的多维连续函数,借助马尔可夫链蒙特卡罗方法对其进行抽样,提高了极大后验的计算速度。实验数据表明,新型协商模型能够提高协商效率,减少协商回合数,并提高协商总体效用。%To improve the effectiveness of automated bilateral multi-issue negotiation,a Bayesian posterior model was set up.It was trained through historical data of negotiations and learned negotiation preference of opponents.Moreover,negotiation counter proposal was suggested according to opponents＇ preference so as to improve the efficiency of negotiations.Hypothesis space was complex multi-dimensions continuous function,Markov Chain Monte Carlo（MCMC） method was used to sample the space.Therefore the computing speed of Maximum a Posterior（MAP） was improved in Bayesian model.Experimental data showed that the proposed model could improve efficiency of negotiation,reduce negotiation rounds and improve the whole negotiation utility.