协同过滤推荐系统中的用户博弈

发布时间：2018-01-02 18:14

本文关键词：协同过滤推荐系统中的用户博弈　出处：《计算机学报》2016年06期 　论文类型：期刊论文

【摘要】：在以协同过滤算法为核心的推荐系统中,一个用户能否获得高质量的推荐不仅取决于用户自身是否积极地参与项目评分,还取决于其他用户是否能提供足够多的评分.由于对项目评分是需要付出成本的,理性的用户总是希望以尽可能少的评分换取高质量的推荐.该文用博弈论的方法对协同过滤系统中的用户评分行为进行分析.考虑到一个用户通常无法观察到其他用户的评分和得到的推荐,该文将用户间的交互建模为不完全信息博弈,并引入"满足均衡"的概念来分析该博弈.该文假定每个用户都对推荐质量有一个预期,当所有用户的预期都得到满足时,博弈即达到均衡.针对所建立的博弈模型,该文设计了一种均衡学习算法,该算法允许用户以逐渐增加评分数量的方式来寻找均衡策略.理论分析和仿真结果均表明,当所有用户对推荐质量有着相似的预期时,所提算法可收敛到满足均衡.这一分析结果可以为协作式系统中激励机制的设计提供启发.
[Abstract]:In the recommendation system with collaborative filtering algorithm as the core, whether a user can get a high quality recommendation depends not only on whether the user himself actively participates in the project score. It also depends on whether other users are able to provide enough ratings, since scoring projects is costly. Rational users always want to exchange as few ratings as possible for high quality recommendations. This paper uses game theory to analyze the behavior of users in collaborative filtering systems, considering that a user is usually unable to observe it. His users' ratings and recommendations. In this paper, the interaction between users is modeled as incomplete information game, and the concept of "satisfying equilibrium" is introduced to analyze the game. It is assumed that each user has an expectation of recommendation quality. When all the users' expectations are satisfied, the game reaches equilibrium. In view of the established game model, this paper designs an equilibrium learning algorithm. The algorithm allows users to search for equalization strategies by gradually increasing the number of ratings. Theoretical analysis and simulation results show that when all users have similar expectations for recommendation quality. The proposed algorithm can converge to meet the equilibrium. This analysis result can provide inspiration for the design of incentive mechanism in cooperative systems.
【作者单位】：清华大学电子工程系;中国传媒大学信息工程学院;
【基金】：国家自然科学基金(61271267,61471025) 高等学校博士学科点专项科研基金(20110002110060)资助
【分类号】：TP391.3
【正文快照】： 法分析了协同过滤推荐系统中的隐私保护问题.在1　引　言他们研究的博弈模型中,用户向推荐服务器提供的评分向量被视为用户的策略.为避免评分数据泄露作为一种有效的信息过滤手段,个性化推荐已隐私,用户可向推荐服务器提供虚假评分,但这会降广泛应用于电子商务、视频分享、在

【相似文献】