融合标签信息和时间效应的矩阵分解推荐算法  

Matrix Decomposition Recommendation Algorithm Combining Label Information and Time Effect

作　　者：赵文涛[1] 任行学 ZHAO Wentao;REN Xingxue(College of Computer Science and Technology,Henan Polytechnic University,Jiaozuo 454000,China)

机构地区：[1]河南理工大学计算机科学与技术学院,河南焦作454000

出　　处：《信息与控制》2020年第4期472-477,488,共7页Information and Control

基　　金：河南省高等学校矿山信息化重点实验室基金资助项目(J1601)。

摘　　要：为了解决过度稀疏的评级矩阵导致矩阵分解中的过拟合问题,提出了一种融合标签和时间信息的矩阵分解推荐模型TTMF(matrix factorization recommendation algorithm fusing tags and time information),以丰富单一数据源,缓解矩阵分解中的过拟合问题.首先通过评级数据和标签信息定义用户标签偏好值和项目标签关联度,分别表征用户对标签的兴趣、标签信息和项目之间的联系,并增加时间信息表示用户兴趣随时间的变化;然后,建立用户—项目、用户—标签和项目—标签矩阵模型,通过梯度下降法进行矩阵分解,完成推荐.基于MovieLens数据集实验结果显示,TTMF算法的RMSE(root mean square error)比传统方法LFM(latent factor model)降低了7%.TTMF算法具有更好的推荐效果.To solve the problem of overfitting caused by excessively sparse rating matrices in matrix decomposition,we propose a matrix factorization recommendation algorithm that fuses tags and time information(TTMF).Labels and time information are combined to enrich a single data source and to mitigate overfitting problems in matrix factorization.First,we define the user's tag preference value and item tag relevance defined by rating data and tag information,which represent the user's interest in tags,the relationship between tag information and items,respectively,and we add time information to indicate changes in user interest over time.The user item,user tag,and item tag matrix models are matrix-decomposed by gradient descent to complete recommendations.Experiment results on the MovieLens dataset show that the root mean square error of the TTMF algorithm is reduced by 7% compared with the traditional method latent factor model,and the TTMF algorithm has a better recommendation effect.

关 键 词：协同过滤算法 矩阵分解 梯度下降法 个性化推荐 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]