融合项目聚类和时间权重的动态协同过滤算法  

Dynamic Collaborative Filtering Algorithm Fusing Item Clustering and Time Weight

作　　者：吴佳炜 沈玲玲[1,2] 钱钢[1] 

机构地区：[1]南京师范大学计算机科学与技术学院,江苏南京210097 [2]南京师范大学商学院,江苏南京210097

出　　处：《南京师范大学学报（工程技术版）》2017年第3期63-69,共7页Journal of Nanjing Normal University(Engineering and Technology Edition)

基　　金：国家自然科学基金(61503188);江苏省自然科学基金(BK20150982)

摘　　要：传统基于项目的协同过滤算法离线计算项目间的相似度,提高了向用户推荐的速度,但极大的数据稀疏度影响了推荐质量,且该算法也忽略了用户兴趣随时间变化这一现象.针对上述问题,提出了一种融合项目聚类和时间权重的动态协同过滤算法,根据用户偏好对项目进行聚类,找出类别偏好相似的候选邻居,再在候选邻居中搜寻最近邻,排除与目标项目共同评分较少的项目干扰,提高了搜寻相似项目的准确性.同时,引入时间权重来反映用户兴趣随时间的变化,从整体上提高推荐质量.在Movie Lens数据集上进行实验,实验结果表明,本文所提出算法的推荐质量较传统的协同过滤算法有显著提高.The traditional item-based collaborative filtering algorithm calculates item-item similarity offline and improves the real-time performance of recommender system,but the big data sparsity problem still impacts the quality of the algorithm and it also ignores the phenomenon that users' interests change over time. To address the issues above,this paper proposes a dynamic collaborative filtering algorithm fusing item clustering and time weight. The proposed algorithm first clusters items according to the user's preference,then finds out candidate neighbors who are similar to the target item in class preference. Then it searches for nearest neighbors in the candidate neighbor set,which eliminates the interference of the items those have few co-ratings with the target item. At the same time,this algorithm introduces time weight to reflect the change of users' interests over time,which improves recommendation quality from the overall. Experimental results based on Movie Lens dataset show that the recommendation quality of the new algorithm is significantly improved compared with traditional item-based collaborative filtering algorithm and user-based collaborative filtering algorithm.

关 键 词：协同过滤 推荐系统 类别偏好 相似度 时间权重 

分 类 号：P208[天文地球—地图制图学与地理信息工程]