摘要
针对传统的相似度计算方法仅依靠用户评分信息矩阵来计算物品或用户相似度,物品相似度的计算考虑了所有用户的历史反馈信息等问题,提出一种改进的协同过滤推荐算法。以所有物品的度的平均值作为阈值,在用户相似度计算公式中引入用户共同评分权重以及流行物品权重;在物品相似度计算公式中引入物品时间差因素和用户共同评分权重。将兴趣相似的用户聚成一类,在类内应用推荐算法分别为用户进行推荐。实验结果表明,相比于传统的协同过滤推荐算法,新算法得到的推荐结果在召回率上提高了2.1%。该算法可在一定程度上提高推荐算法的精度以及推荐质量。
There exists several issues in traditional collaborative filtering algorithms:a)It takes the impact of all users’histori-cal feedback information into account when calculating the similarities between any two items;b)It only utilizes the user’s rating data when calculating the similarities.However,the user group that has similar interests with the target user has a higher reference value than other users.Considering the fact that irrelevant historical information leaded to poor recommendation results,this paper proposed a novel collaborative filtering recommendation algorithm based on K-means clustering.The new algorithm refined the user’s similarity metric with the user’s common rating weight and popular items weight,the item’s similarity metric with time difference weight and user’s rating weight respectively,and clustered all uses into several partitions according to the similarities.Then,it applied recommend algorithm in each of the clusters.Experimental results show that,compared with traditional item-based top-N collaborative filtering recommendation algorithm,the proposed algorithm can improve the recall by 2.1%on average.The proposed algorithm can improve the accuracy and the quality of the recommendation effectively.
作者
肖文强
姚世军
吴善明
Xiao Wenqiang;Yao Shijun;Wu Shanming(College of Science,Information Engineering University,Zhengzhou 450001,China)
出处
《计算机应用研究》
CSCD
北大核心
2018年第1期105-108,112,共5页
Application Research of Computers
关键词
协同过滤推荐算法
用户评分信息
相似度
聚类算法
召回率
collaborative filtering recommendation algorithm
wser’s rating information
similarity
clustering algorithm
recall