摘要
为了降低数据稀疏性的影响,提高推荐系统的推荐生成质量,提出了一种基于多层相似性用户聚类的协同过滤推荐算法。该算法采用新的多层用户相似性度量,并将推荐过程分成了离线和在线两个部分。离线时,算法对基本用户数据进行预处理,并对基本用户聚类;在线时,算法利用已有的用户聚类寻找目标用户最近邻居,并产生推荐。实验表明,该算法不仅加快了推荐生成速度,而且提高了推荐质量,降低了约6%的平均绝对误差。
To overcome the difficulty of data sparsity in recommendation systems, a collaborative filtering （CF） algorithm based on clustering basal users is presented. The algorithm uses a new measurement of multiple-level similarities between the basal users and separates the procedure of recommendation into offline and online phases. In the offline phase, the data of basal users are preproeessed, and the basal users are clustered. Then, in the online phase, the nearest neighbors of an active user are found according to the basal user clusters, and the recommendation to the active user is produced. Experimental results show that the algorithm improves the performance of CF systems in both the recommendation quality and the efficiency, and decreases the mean absolute error about 6%.
出处
《南京航空航天大学学报》
EI
CAS
CSCD
北大核心
2006年第6期717-721,共5页
Journal of Nanjing University of Aeronautics & Astronautics
基金
江苏省自然科学基金(BK2002091)资助项目
关键词
推荐算法
协同过滤
聚类
平均绝对误差
recommendation algorithm
collaborative filtering （CF）
cluster
mean absolute error （MAE）
