基于推荐质量的信任感知推荐系统

Quality of Recommendation Based Trust-aware Recommender System

推荐系统在解决信息过载方面已经取得了很大的成功,同时也存在数据稀疏、冷启动等问题。如何在评分数据稀疏的情况下获得满意的推荐成为推荐系统亟待解决的问题。将信任引入推荐系统成为解决上述问题的有效方法之一。已有的信任感知推荐系统大多基于布尔型信任关系,且没有考虑信任的领域相关性。在服务选择领域,服务请求者依据QoS(quality of service)选择服务。联想到在服务推荐领域推荐请求者可以依据推荐质量(quality of recommendation,QoR)选择推荐用户,提出了推荐质量(QoR)的概念和基于推荐质量的信任感知推荐系统。QoR的属性包含评价相似度、领域信任值、领域相关度和亲密程度,利用信息熵方法可确定各属性的权重。仿真表明该方法提高了推荐系统在数据稀疏情况下的精确度和评分覆盖率,有效提高了冷启动用户的召回率,在一定程度上解决了冷启动问题。

Recommender system has achieved great success in dealing with information overload,meanwhile has some problems,such as data sparse,cold start and so on.How to get satisfied recommendation under the circumstance of data sparse is urgent for recommender system.Introducing trust into recommender system is an efficient way to resolve the above problems.Most of existing trust-aware recommender systems are based on Boolean trust relationship,and do not take domain correlation of trust into account.In the field of service selection,service requestor selects services based on QoS（quality of service）.Inspired by QoS,recommendation requestor can find recommender based on QoR（quality of recommendation）.Thus we put forward the concept of QoR and QoR based trust-aware recommender system.The attributes of qor include user rating similarity,domain trust,domain relative degree and social intimacy degree,whose weights are determined by method of information entropy.Empirical evaluation shows that our method improves the precision and ra-ting coverage of recommender system under condition of data sparse,above all,effectively improves the recall rate of cold start user,resolving the cold start problem to some extend.