基于Nearest-Biclusters协作过滤技术的效用图结构学习算法

An Algorithms for Learning the Structure of Utility Graph Based on Nearest-Biclusters Collaborative Filtering

在多议题协商研究中,议题之间的依赖关系增加了协商Agent效用函数的复杂性,从而使得多议题协商变得更加困难.基于效用图的多议题依赖协商模型是体现议题间依赖关系的多议题协商模型.在该协商模型中,协商双方仅需要较少的协商步数就能够找到满足Pareto效率的协商结局.如何有效地学习买方Agent的效用图结构是该协商模型的关键.文中基于Nearest-Biclusters协作过滤技术的思想提出了一种新的效用图结构学习算法(NBCFL算法).该算法首先利用Nearest-Biclusters协作过滤技术发现买方偏好的局部匹配特性,提取与当前买方Agent类型相同的买方Agent所产生的协商历史记录,然后通过计算各议题间的依赖度学习买方Agent的效用图结构.实验表明在参与协商的买方Agent类型不同的条件下,NBCFL算法比IBCFL算法能更好地学习买方Agent的效用图结构.

In the research of multi-issue negotiation,the interdependencies between issues greatly complicates the negotiation agents＇ utility functions,so this makes negotiation more difficultly.The multi-issue negotiation model based on Utility Graph is the multi-issue dependence negotiation model which considers interdependencies between issues.The negotiants need a few number of negotiation steps to reach Pareto-efficient agreements in the negotiation model.The key problem of the negotiation model is how to learn the structure of Utility Graph effectively.This paper proposes a new algorithm for learning the structure of Utility Graph based on Nearest-Biclusters Collaborative Filtering（NBCFL）.Firstly,the algorithm takes advantage of the trait that Nearest-Biclusters Collaborative Filtering could detect partial matching of buyers＇ preferences,and collects the negotiation data which were produced by buyer that is the same class with active buyer.Secondly,it retrieves the structure of Utility Graph using the degree of interdependencies between issues.The experiments show that NBCFL algorithm can learn the structure of Utility Graph more effectively than IBCFL algorithm on condition that buyers from different classes of buyers.