摘要
针对传统的协同过滤(CF)方法由于仅利用评分数据而存在用户偏好挖掘不全面,以及数据稀疏性及冷启动问题,提出了一种基于softmax回归和矩阵分解的协同过滤推荐方法(SRMF-based CF)。该方法首先利用文本挖掘技术,从文本评论中提取项目特征并进行基于特征的情感强度分析,然后采用softmax回归和矩阵分解技术预测目标用户在特征层面的情感值并预测该用户对未评分项目的总体评分值。真实数据集的实验表明,与传统的基于总体评分的协同过滤、基于文本评论的协同过滤以及基于矩阵分解的协同过滤方法相比,SRMF-based CF在平均绝对误差(MAE)和均方根误差(RMSE)上分别降低了1.8%~7.6%和2.1%~9.7%,取得了较好的推荐效果。
Concerning the incomplete mining of user preferences,data sparsity and cold start of traditional collaborative filtering methods,a Collaborative Filtering recommendation method based on Softmax Regression and Matrix Factorization(SRMF-based CF)was proposed.Firstly,features were extracted from textual comments by text mining technology and feature-based sentiment intensity analysis was conducted.Then,softmax regression and matrix factorization techniques were used to predict the target users sentiment intensity values at the feature level.Based on this,the overall ratings that the users have not evaluated were predicted.Experiments on real dataset show that compared with traditional collaborative filtering methods(collaborative filtering based on overall ratings,collaborative filtering based on textual reviews,and collaborative filtering based on matrix factorization),the proposed SRMF-based CF decreases Mean Absolute Error(MAE)and Root Mean Squared Error(RMSE)by 1.8%to 7.6%and 2.1%to 9.7%,respectively,achieving better recommendation performance.
作者
王努努
WANG Nunu(School of Business,Central South University,Changsha Hunan 410083,China)
出处
《计算机应用》
CSCD
北大核心
2019年第S02期127-131,共5页
journal of Computer Applications
基金
中南大学中央高校基本科研业务费专项资金资助项目(2018zzts299)
关键词
文本评论
推荐方法
协同过滤
softmax回归
矩阵分解
textual review
recommendation method
Collaborative Filtering(CF)
softmax regression
matrix factorization
