结合隐式反馈与相关性建模的概率流式张量分解推荐模型  

Probabilistic streaming tensor decomposition recommendation model with implicit feedback and correlation modeling

作　　者：赵建立 姚彬 陈建建 李浩 ZHAO Jianli;YAO Bin;CHEN Jianjian;LI Hao(College of Computer Science and Engineering,Shandong University of Science and Technology,Qingdao 266590,China)

机构地区：[1]山东科技大学计算机科学与工程学院,山东青岛266590

出　　处：《山东科技大学学报（自然科学版）》2023年第3期76-84,共9页Journal of Shandong University of Science and Technology(Natural Science)

基　　金：国家自然科学基金项目(62072288);山东省自然科学基金项目(ZR2021MF104,ZR2021MF113);青岛市科技计划重点研发专项(21-1-2-19-xx);青岛市西海岸新区科技计划项目(2020-1-6)。

摘　　要：在推荐系统中,流式张量分解模型常被用于分析和处理高维流式数据,并向用户推荐可能感兴趣的商品。然而现有模型常常忽略对象特征之间的相关性,缺乏有效的先验建模,并且模型大多仅使用显式评分数据,忽略对隐式反馈信息的利用,导致模型精度下降。针对以上问题,提出一种结合隐式反馈与相关性建模的概率流式张量分解推荐模型(POSTFR)。POSTFR通过添加具有Lewandowski-Kurowicka-Joe(LKJ)先验的协方差矩阵建模对象间的相关性,并在其中加入隐式反馈信息以提高频繁出现的对象的权重,提高了模型的准确度。在多个真实数据集上,POSTFR比其他概率流式模型取得更好的效果。These applications generate streaming data based on users’interactive behavior.In recommender systems,streaming tensor decomposition models are often employed for analyzing and processing streaming data,and for re-commending items that users may be interested in.However,existing models often neglect the correlation between object features and lack effective prior modeling.Besides,most models only use explicit rating data without the use of implicit feedback information,thus exacerbating the accuracy loss of models.In this paper,we proposed a probabilistic streaming tensor decomposition recommendation model with implicit feedback and relationship modeling(POSTFR).POSTFR modeled the relationship between objects by adding covariance matrices with Lewandowski-Kurowicka-Joe(LKJ)priors and incorporated implicit feedback information in the updating processto increase the weight of frequently occurring objects.In this way,the accuracy of the proposed model was improved.Compared to other probabilistic streaming models,POSTFR obtained better results in several real datasets.

关 键 词：张量分解 流式数据 隐式反馈 概率流式模型 推荐模型 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]