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.
User participation,i.e.providing ratings to the recommendation server,is of vital importance for the success of collaborative filtering-based recommendation systems.As the name collaborative suggests,whether a user can get high-quality recommendations depends not only on the user himself/herself but also on other users.However,due to the rating cost,rational users prefer to provide as few ratings as possible.In this paper,we model the interactions among users as a game with incomplete information and apply the notion of satisfaction equilibrium（SE）to the proposed game.Every user is assumed to have an expectation for the recommendation quality,and when all users＇ expectations are satisfied,a SE of the game is achieved.We design a behavior rule which allows users to achieve a SE via iteratively rating items.Theoretical analysis and simulation results demonstrate that,if all users have moderate expectations for recommendation quality and satisfied users are willing to provide more ratings,then all users can get satisfying recommendations without providing many ratings.We hope the game analysis presented in this paper can provide some implications for designing mechanisms to encourage user participation.
出处
《计算机学报》
EI
CSCD
北大核心
2016年第6期1176-1189,共14页
Chinese Journal of Computers
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
关键词
协同过滤
博弈论
满足均衡
均衡学习
收敛条件
社交网络
社会媒体
collaborative filtering
game theory
satisfaction equilibrium
equilibrium learning
convergence condition
social network
social media
