基于项目近邻的约束概率矩阵分解算法  

A Constrained Probabilistic Matrix Factorization Algorithm Based on Item-neighborhood

作　　者：张天杰[1] 曹苏燕[1] 闫世洋 

机构地区：[1]南京邮电大学计算机学院,江苏南京210003

出　　处：《计算机技术与发展》2016年第10期64-68,共5页Computer Technology and Development

基　　金：国家"863"高技术发展计划项目(2006AA01Z201)

摘　　要：在个性化推荐领域,协同过滤是目前最为成功应用最为广泛的推荐技术之一。约束概率矩阵分解算法便是一种基于模型的协同过滤算法,它能有效地面对推荐系统中遇到的海量数据问题,保证推荐的实时性。然而,传统的约束概率矩阵分解算法并没有考虑用户或者项目之间的关系,使得算法的推荐质量受到影响。为进一步提高算法推荐的质量,文中在约束概率矩阵分解算法模型的基础上引入项目近邻关系,通过结合从项目简介中提取的固有特征和用户评定的标签特征两方面信息来确定项目的最近邻居集合,并将该邻居集合融合到基于约束的概率矩阵分解模型中产生推荐。通过在真实的数据集上的验证结果表明,该算法能够更有效地预测用户对项目的评分,提高算法的推荐精度。Collaborative filtering is one of the most successful applications in the field of personalized recommendation. Constrained probabilistic matrix factorization is a model-based collaborative filtering algorithm which can effectively deal with the problem of scaiability in large-scale recommendation system and guarantee the real-time of recommendation. However, the traditional one does not consider the relationship between the users or the items,which makes the quality of the algorithm affected. It takes the relationship of item-neighbor- hood into the aigorithm model of constrained probability matrix factorization to improve the quality of the proposed algorithm. To guarantee the accuracy of item-neighborhood,the inherent features extracted from the item＇ s summary and the tag of user marked for the item are used to get the set of the nearest neighbor for the items, then the item-neighborhood set is applied into the framework of the constrained probabilistic matrix factorization algorithm. The experiments on real datasets show that the proposed algorithm can predict the user＇ s rating on the item more effectively ,and improve the accuracy of the recommendation.

关 键 词：推荐系统 协同过滤 约束概率矩阵分解 项目近邻 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]