基于用户属性和生成对抗网络的推荐系统  

Recommendation system based on users’attributes and generative adversarial networks

作　　者：王永强 陈徐洪 张壮壮 董云泉 WANG Yong-qiang;CHEN Xu-hong;ZHANG Zhuang-zhuang;DONG Yun-quan(School of Electronic and Information Engineering,Nanjing University of Information Science and Technology,Nanjing 210044,China;Chongqing Branch,China Development Bank,Chongqing 400023,China)

机构地区：[1]南京信息工程大学电子与信息工程学院,江苏南京210044 [2]国家开发银行重庆市分行,重庆400023

出　　处：《计算机工程与设计》2024年第1期275-281,共7页Computer Engineering and Design

基　　金：国家自然科学基金面上基金项目(62071237)。

摘　　要：为提升推荐精度,解决传统推荐算法在用户评分向量中存在的未评分项语义模糊造成的推荐精度下降问题,提出一种基于用户属性的条件生成对抗网络的推荐方法。将用户的属性特征进行提取和编码,并作为生成对抗网络的条件,通过这种明确信号指导用户偏好的生成并进行推荐。在两个公开的电影评分数据集上进行实验,实验结果表明,所提方法可以有效改善推荐精度,在各评价指标上均优于现有方法,具有一定实用价值。To improve the recommendation accuracy and solve the problem of recommendation accuracy decline caused by the semantic ambiguity of unrated items in the user rating vector of traditional recommendation algorithms,a recommendation method of generative adversarial networks(GAN)based on users’attributes was proposed,in which users’attributes were extracted and encoded as the conditions of GAN,and the explicit signals were used to guide the generation and recommendation of user prefe-rences.Results of experiments on two public movie rating datasets show that the proposed method can effectively improve the recommendation accuracy,and it is superior to the existing methods in various evaluation indexes with certain practical value.

关 键 词：推荐系统 生成对抗网络 用户属性 协同过滤 评分矩阵 特征提取 梯度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]