摘要
现有的推荐算法主要依靠评分记录,对用户的个性需求关注较少,推荐结果不完全符合实际需求。针对该问题,本文在传统的基于用户的协同过滤算法(UCF)基础上,结合密度峰值聚类研究物品属性,分析用户对物品类型、聚类的兴趣取向,深入挖掘用户的个性需求,提出了一种结合物品类型和密度峰值聚类的协同过滤推荐算法。采用密度峰值聚类,无需指定聚类中心和聚类数,利用逆向文档频率对算法进行优化,提高了对物品特征和用户兴趣的识别度。实验结果表明,本文算法能较好地获取用户偏向,提供更加准确、高效的Top-N推荐。
Most of existing recommendation algorithms depend on items’rating records marked by users and pay less attention to users’individual requirements such that the recommendation results may not really conform to the practical demands.Aiming at the above shortcoming,by integrating genre and density-peaks-clustering technique,this paper proposes an improved user-based collaborative filtering recommendation algorithm for meeting various individual needs of users.Firstly,the users’interests in different genres are analyzed according to historical rating data and the genre of items.And then,the density peaks clustering algorithm is utilized to research items’attributes and analyze users’interests in different clusters of items,by which the users’individual demand can be dug deeply.Since the proposed method uses density peaks clustering,it doesn’t need to assign the centers of clustering uses and specify the quantity of clusters before clustering,which can increase the accuracy of clustering.Moreover,this paper uses inverse document frequency to optimize the proposed algorithm such that the identification degree of items’characteristics and users’interests can be effectively improved.Finally,it is shown from the experiment with the movielens dataset that the proposed algorithm can get users’interests effectively and provide more accurate and efficient Top-N recommendation.
作者
陈帆
孙自强
CHEN Fan;SUN Zi-qiang(Key Laboratory of Advanced Control and Optimization for Chemical Process,Ministry of Education,East China University of Science and Technology,Shanghai 200237,China)
出处
《华东理工大学学报（自然科学版）》
CAS
CSCD
北大核心
2018年第6期862-868,共7页
Journal of East China University of Science and Technology
基金
中央高校基本科研业务费重点科研基地创新基金(222201717006)
关键词
密度峰值聚类
协同过滤
推荐算法
density peaks clustering
collaborative filtering
recommendation algorithm
