摘要
传统IPTV推荐难以挖掘用户与节目之间隐式反馈,同时面向实践的IPTV推荐引擎单一。提出基于图表示学习的IPTV推荐系统,构建“用户—节目”历史交互为高阶异构图;利用轻量图卷积操作捕获用户与节目之间的协同信号提取用户偏好,进行评分预测。与MostPopular、BPR、基于神经网络和基于传统图神经网络的协同过滤等主流技术对比验证了方法的先进性,并利用某通信运营商脱敏数据分析验证了在实际场景中的可用性。
Traditional IPTV recommendation is difficult to mine the implicit feedback between users and programs,and the practice-oriented IPTV recommendation engine is single.An IPTV recommendation system based on graph learning is proposed,and the historical interaction of"User-Program"is constructed as a high-order heterogeneous graph.The lightweight graph convolution operation is used to capture the collaborative signal between users and programs to achieve rating prediction.Comparing with mainstream technologies such as MostPopular,BPR,collaborative filtering based on neural network and traditional graph neural network,the advanced nature of the method is verified,and the desensitization data analysis of a communication operator is used to verify the usability in actual scenarios.
作者
危枫
胡飞
王晨子
王丽平
杨佳佳
杨正益
WEI Feng;HU Fei;WANG Chenzi;WANG Liping;YANG Jiajia;YANG Zhengyi(China Telecom Corporation Limited Chongqing Branch,Chongqing 401121;School of Big Data&Software Engineering,Chongqing University,Chongqing 401331)
出处
《软件》
2022年第6期6-8,25,共4页
Software
基金
移动群智感知环境下可信共享和异构数据融合的服务推荐研究(62072060)。
关键词
IPTV
推荐系统
图卷积神经网络
IPTV
recommendation system
graph convolutional neural network