协同过滤推荐算法中的相似性度量研究  

Research on Similarity Measure in Collaborative Filtering Recommendation Algorithm

作　　者：李散散 陈小荣 LI Sansan;CHEN Xiaorong(School of Engineering,Guangzhou College of Technology and Business,Guangzhou 510850,China)

机构地区：[1]广州工商学院工学院,广东广州510850

出　　处：《现代信息科技》2022年第15期59-62,66,共5页Modern Information Technology

基　　金：2021年广州工商学院质量工程建设项目(ZC20211129)。

摘　　要：协同过滤算法作为最古老的算法有着相当广泛的应用,相似度的计算和最近邻居的选择是该算法的核心。在阐述协同过滤推荐算法的原理和常见相似度计算方法的基础上,提出改进的相似度计算方法,并通过实验验证了不同相似度计算方法在推荐效果方面的差异,分析了如何解决数据稀疏性和平衡项目本身质量权重的问题。实验结果表明,改进的相似度计算方法在准确率、召回率、RMSE、MAE四个评估指标上都有更好的表现,因此,该方法能够提高推荐质量。As the oldest algorithm,Collaborative Filtering algorithm has a wide range of applications.The calculation of similarity and the selection of nearest neighbor are the core of the algorithm.Based on explaining the principle of Collaborative Filtering recommendation algorithm and common similarity calculation methods,this paper proposes an improved similarity calculation method,and verifies the differences in recommendation effects of different similarity calculation methods through experiments,and analyzes how to solve the problem of data sparsity and balance the quality weight of the project itself.The experimental results show that the improved similarity calculation method performs better on four evaluation indicators:accuracy rate,recall rate,RMSE and MAE.Therefore,the method can improve the recommendation quality.

关 键 词：协同过滤 推荐算法 相似度 最近邻居 数据稀疏性 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391[自动化与计算机技术—控制科学与工程]