一种改进相似度计算方法的协同过滤推荐算法  

An improved collaborative filtering recommendation algorithm based on an improved similarity calculation method

作　　者：祝奇伟 陈家琪[1] 

机构地区：[1]上海理工大学计算机科学与工程系,上海200093

出　　处：《信息技术》2015年第3期13-16,共4页Information Technology

基　　金：上海市教委科研创新基金(12zz146)

摘　　要：由于新用户和新项目的不断加入,传统的协同过滤推荐算法存在冷启动问题。针对该问题,提出了一种改进相似度计算方法的协同过滤推荐算法。首先根据项目的属性特征,计算项目的属性相似性,然后根据项目的用户评分,计算项目的得分相似性,按一定的权重比例将两种相似性组合起来作为最终的项目相似性。最后,根据项目相似性计算目标项目的邻居项目集,根据邻居项目集预测目标项目的用户评分。实验结果表明,新算法能提高推荐精度,并能在一定程度上解决冷启动问题。Due to the new users and new items,cold start issue exists in the traditional collaborative filtering recommendation algorithm. To address this problem, a new collaborative filtering recommendation algorithm based on the new similarity calculation method is proposed. Firstly,it uses the characters of item attributes to compute attribute similarity of the items. Secondly,it uses the rating records to compute item score similarity. In order to get the comprehensive project similarity,combines attribute similarity and project score similarity with weighting factors. Then it gets the neighbors of the target item according to the comprehensive item similarity. At last,it predicts the rating of the target item according the neighbors. The experimental results show that the new algorithm can improve the precision of recommendation and solve cold start issue.

关 键 词：协同过滤 推荐 相似度 组合 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]