基于时序行为的协同过滤推荐算法  

Recommendations Based on Collaborative Filtering by Exploiting Sequential Behaviors

作　　者：孙光福[1] 吴乐[1] 刘淇[1] 朱琛[1] 陈恩红[1] 

机构地区：[1]中国科学技术大学计算机科学与技术学院,安徽合肥230027

出　　处：《软件学报》2013年第11期2721-2733,共13页Journal of Software

基　　金：国家自然科学基金(61073110);国家科技支撑计划(2012BAH17B03)

摘　　要：协同过滤直接根据用户的行为记录去预测其可能喜欢的产品,是现今最为成功、应用最广泛的推荐方法.概率矩阵分解算法是一类重要的协同过滤方式.它通过学习低维的近似矩阵进行推荐,能够有效处理海量数据.然而,传统的概率矩阵分解方法往往忽略了用户(产品)之间的结构关系,影响推荐算法的效果.通过衡量用户(产品)之间的关系寻找相似的邻居用户(产品),可以更准确地识别用户的个人兴趣,从而有效提高协同过滤推荐精度.为此,提出一种对用户(产品)间的时序行为建模的方法.基于该方法,可以发现对当前用户(产品)影响最大的邻居集合.进一步地,将该邻居集合成功融合到基于概率矩阵分解的协同过滤推荐算法中.在两个真实数据集上的验证结果表明,所提出的SequentialMF推荐算法与传统的使用社交网络信息与标签信息的推荐算法相比,能够更有效地预测用户实际评分,提升推荐精度.Collaborative filtering, which makes personalized predictions by learning the historical behaviors of users, is widely used in recommender systems. The key to enhance the performance of collaborative filtering is to precisely learn the interests of the active users by exploiting the relationships among users and items. Though various works have targeted on this goal, few have noticed the sequential correlations among users and items. In this paper, a method is proposed to capture the sequential behaviors of users and items, which can help find the set of neighbors that are most influential to the given users （items）. Furthermore, those influential neighbors are successfully applied into the recommendation process based on probabilistic matrix factorization. The extensive experiments on two real-world data sets demonstrate that the proposed SequentialMF algorithm can achieve more accurate rating predictions than the conventional methods using either social relations or tagging information.

关 键 词：协同过滤 时序行为 概率矩阵分解 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]