音乐个性化推荐算法TFPMF的研究  

Personalized Music Recommendation Algorithm TFPMF

作　　者：叶西宁[1] 王猛[1] Ye Xining;Wang Meng(College of Information Science and Engineering,East China University of Science and Technology,Shanghai 200237,China)

机构地区：[1]华东理工大学信息科学与工程学院

出　　处：《系统仿真学报》2019年第7期1397-1407,共11页Journal of System Simulation

基　　金：国家自然科学基金(61304071)

摘　　要：基于情境感知的个性化推荐是近年来推荐系统中的研究热点和难点问题,数据稀疏是当前推荐系统面临的主要问题。以音乐推荐为背景,改进了多种情境信息的表示方法,将优化排名倒数(RR)的概率矩阵分解模型(RR-PMF)与张量分解相结合,提出了张量概率矩阵分解模型(TFPMF),并使用交叉最小二乘法(ALS)优化该模型。使用last.fm数据集进行仿真实验,通过仿真模型得出TOP-N推荐列表,结果表明该算法在准确率(Precision)、召回率(Recall)和标准化折算累加值(NDCG)评价指标上具有很大的优势,该算法能够有效缓解数据稀疏问题。In recent years, the personalized context-aware recommendation is the rub and hotness in the research of recommendation system, and the data sparseness is the main problem faced by the current recommendation systems. In the setting of music recommendation, the representing method of varieties of situational information is improved. A model of TFPMF is proposed, which combines the model of RR-PMF with the tensor decomposition. TFPMF is optimized by alternative least squares(ALS). By the simulation experiments in the last.fm dataset, we got the TOP-N recommended list through the simulation program. The simulation results show that the proposed algorithm has great advantages in the evaluation index of Precision, Recall and NDCG, and the algorithm can effectively alleviate the data sparsity problem.

关 键 词：推荐系统 排名倒数 概率矩阵分解 张量分解 TFPMF(基于张量分解的概率矩阵分解) 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]