基于矩阵填充和物品可预测性的协同过滤算法  

Collaborative Filtering Recommendation Algorithm Based on Rating Matrix Filling and Item Predictability

作　　者：潘涛涛[1] 文锋[2] 刘勤让[1] 

机构地区：[1]国家数字交换系统工程技术研究中心,郑州450002 [2]江南计算技术研究所,无锡214000

出　　处：《自动化学报》2017年第9期1597-1606,共10页Acta Automatica Sinica

基　　金：国家高技术研究发展计划(863计划)(2014AA01A);国家自然科学基金(61572520)资助~~

摘　　要：针对传统矩阵填充算法忽略了预测评分与真实评分之间的可信度差异和传统Top-N方法推荐精度低等问题,提出了一种改进的协同过滤算法.该算法首先利用置信系数C区分评分值之间的可信度;然后提出物品可预测性的概念,综合物品的预测评分与物品的可预测性进行物品推荐并将其转化为0-1背包问题,从而筛选出最优化的推荐列表.实验结果表明:该算法能有效缓解稀疏性的影响,提高推荐性能,并且算法具有良好的可扩展性.The traditional matrix filling algorithm ignores the difference between true rating and predictive rating, and there is only one standard on the traditional Top-N recommended method. In order to solve these two problems, an im- proved collaborative filtering algorithm is proposed. Firstly, the confidence coefficient is used to distinguish the credibility of the ratings. Then, a concept of item predictability is proposed. The program recommends items by comprehensively considering the item＇s predictive ratings and the predictability, and transforming the program into the 0-1 knapsack problem so as to select the optimized recommended list. Experimental results show that the algorithm can effectively alleviate the effect of sparsity and improve the performance of the recommendation, and that the optimization algorithm has good expansibility.

关 键 词：协同过滤 推荐系统 预测评分 相似度 0-1背包问题 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]