基于MMTD和兴趣偏向系数的协同过滤推荐算法  

Collaborative filtering recommendation algorithm based on MMTD and interest bias coefficient

作　　者：陆荣 周宁宁[1] Lu Rong;Zhou Ningning(School of Computer,Nanjing University of Posts&Telecommunications,Nanjing 210023,China)

机构地区：[1]南京邮电大学计算机学院,南京210023

出　　处：《计算机应用研究》2020年第9期2600-2603,共4页Application Research of Computers

基　　金：智能电网保护和运行控制国家重点实验室开放课题(20169,201610);国家自然科学基金资助项目(61170322,61373065,61302157)。

摘　　要：针对传统基于用户的协同过滤推荐算法存在的相似性度量不准确和缺乏对用户评分合理应用的问题,提出了一种结合中介真值程度度量(MMTD)和兴趣偏向系数的推荐算法。该方法首先采用MMTD度量用户评分的相似性;然后利用用户评分相似性改进余弦相似性公式和Jaccard公式,得到新的基于MMTD的用户相似性度量方法;最后结合兴趣偏向系数输出推荐结果。在MovieLens-100k数据集上的实验结果表明,该方法可以在一定程度上提高用户间相似性度量的准确性,提高推荐结果的准确率和召回率。Due to the inaccuracy of the traditional user-based collaborative filtering recommendation algorithm and the lack of reasonable application of user ratings,this paper proposed a recommendation algorithm combining the measure of medium truth degree( MMTD) and the interest bias coefficient. The method firstly used MMTD to measure the similarity of user ratings.Secondly,it improved the cosine similarity formula and Jaccard formula by adopting user score similarity,and obtained a new MMTD-based user similarity measurement method. Finally,combining the interest bias coefficient,it output the recommendation result. The experimental results on the MovieLens-100 k dataset show that the algorithm can improve the accuracy of the similarity measure between users to some extent,and improve the precision and recall of recommendation results.

关 键 词：协同过滤 用户评分 用户相似度 中介真值程度度量 兴趣偏向系数 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]