基于用户行为数据的非负矩阵分解音乐软件推荐算法研究  

Research on Non-negative Matrix Factorization Music Software Recommendation Algorithm Based on User Behavior Data

作　　者：金龙 JIN Long(Faculty of Engineering,The University of Sydney,New South Wales 2006,Australia)

机构地区：[1]悉尼大学工程学院,新南威尔士州2006

出　　处：《现代信息科技》2025年第8期111-116,共6页Modern Information Technology

摘　　要：随着互联网音乐服务的普及,如何为用户精准推荐音乐成为一个重要的研究课题。文章针对现有音乐推荐系统在处理冷启动、数据稀疏等问题时的不足,提出了一种基于非负矩阵分解(NMF)的音乐推荐算法。研究使用了与网易云音乐合作项目的数据集,该数据集包含超过200万用户的5 700多万条音乐交互记录。通过引入用户行为权重和稀疏约束,分别构建了加权NMF和稀疏NMF模型。实验结果表明,加权NMF在处理高频交互用户时表现最佳,F1值达到0.997 6;而稀疏NMF在处理冷启动用户时更具优势,对于交互次数少于10次的用户,其推荐准确率比基础NMF提升了15%。研究成果为音乐推荐系统的优化提供了新的解决方案。With the popularity of internet music services,how to accurately recommend music for users has become an important research topic.This paper aims at the shortcomings of the existing music recommendation system in dealing with problems such as cold-start and data sparsity.A music recommendation algorithm based on Non-Negative Matrix Factorization(NMF)is proposed.The study uses a dataset from a collaboration project with NetEase Cloud Music,which contains more than 57 million music interaction records of more than 2 million users.By introducing user behavior weights and sparse constraints,weighted NMF and sparse NMF models are constructed respectively.The experimental results show that the weighted NMF performs best when dealing with high-frequency interactive users,and the F1 score reaches 0.9976.The sparse NMF has more advantages in dealing with cold-start users.For users with fewer than 10 interactions,the recommendation accuracy is 15%higher than that of the basic NMF.The research results provide new solutions for the optimization of the music recommendation system.

关 键 词：机器学习 音乐推荐模型 NMF 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]