融合基础属性和通信行为的移动用户个性化推荐  

Personalized recommendation of mobile users by integrating basic information and communication behavior

作　　者：吴贤君 唐绍诗 王明秋[2] WU Xianjun;TANG Shaoshi;WANG Mingqiu(School of Statistics and Mathematics,Zhongnan University of Economics and Law,Wuhan 430073,Hubei,China;School of Statistics and Data Science,Qufu Normal University,Jining 273165,Shandong,China)

机构地区：[1]中南财经政法大学统计与数学学院,湖北武汉430073 [2]曲阜师范大学统计与数据科学学院,山东济宁273165

出　　处：《山东大学学报（理学版）》2023年第9期81-93,共13页Journal of Shandong University(Natural Science)

基　　金：全国统计科学研究重点项目(2020LZ26,2022LY071)。

摘　　要：基于经典矩阵分解模型,提出融合用户基础属性和通信行为的矩阵分解模型,对比两种模型在评分预测和Top-N推荐问题上的表现。对评分预测问题采用RFM模型构造用户-产品评分矩阵,并结合移动产品的特点对RFM模型中部分指标进行调整,得到能够更加准确、客观地反映用户对产品兴趣偏好的评分矩阵。对Top-N推荐问题采用将用户未有过订购行为的热门产品优先纳入负样本的负采样方法。结果表明,融合用户基础属性和通信行为的矩阵分解模型在两种问题上具有更好的表现。Considering the factors which affect the mobile usersproduct ordering,this paper proposes a new matrix factorization model integrating the basic information and communication behavior of the mobile users,and compares the performance of the proposed method with the traditional model on the prediction of rating and Top-N recommendation.For the prediction of rating,some indices are adjusted in the RFM model by combining with the characteristics of mobile users in the product ordering process.Then a user-product rating matrix,constructed by the adjusted RFM model,can accurately and objectively reflect the userspreference for products.For the Top-N recommendation,a negative sampling method is adopted that popular products with no ordering behavior are preferred to be included in the negative samples.The numerical results show that the proposed method performs better.

关 键 词：移动用户 矩阵分解模型 RFM模型 评分预测 Top-N推荐 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]