基于深度神经网络融合特征信息的协同过滤模型  

Deep collaborative filtering based on two-tower structure fusion attribute information

作　　者：张铁斌 郑行 邢星 陈建 贾志淳[2] ZHANG Tiebin;ZHENG Hang;XING Xing;CHEN Jian;JIA Zhichun(Information Construction and Management Center,Bohai University,Jinzhou 121013,China;College of Information Science and Technology,Bohai University,Jinzhou 121013,China)

机构地区：[1]渤海大学信息化建设与管理中心,辽宁锦州121013 [2]渤海大学信息科学与技术学院,辽宁锦州121013

出　　处：《渤海大学学报（自然科学版）》2022年第2期172-177,共6页Journal of Bohai University:Natural Science Edition

基　　金：国家自然科学基金项目(No:61972053);辽宁省教育厅科学研究项目(No:LQ2019016;No:LJ2019015);辽宁省自然科学基金项目(No:2019-ZD-0505);辽宁“百千万人才工程”A类项目培养经费资助(No:2021921024).

摘　　要：针对神经协同过滤方法仅对用户和物品的交互信息建模,无法学习用户和项目辅助信息中的潜在特征的问题,提出一种基于深度神经网络融合辅助信息的协同过滤模型.该模型用双重的多层感知机分别对用户和项目的交互信息以及辅助信息建模,计算用户对物品感兴趣的概率.在MovieLens和The Movies数据集上的实验表明,本文模型在PR-AUC高于以往的协同过滤算法,验证了该模型在推荐系统中的有效性和可行性.In order to solve the problem that the neural collaborative filtering method only models the interaction information between users and items and cannot learn the potential features of the auxiliary information between users and items,a collaborative filtering model based on deep neural network fusion of auxiliary information was proposed.The model uses dual multi-layer perceptron to model the interactive information and auxiliary information of users and items,respectively,and calculates the probability of users′interest in items.Experiments on MovieLens and The Movies data sets show that the PR-AUC performance of the proposed model is higher than the previous collaborative filtering algorithms,which verifies the effectiveness and feasibility of the proposed model in recommendation systems.

关 键 词：神经网络 深度学习 协同过滤 隐性反馈 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]