基于图神经网络的视频推荐系统

2021-02-07 作者:高宸,李勇,金德鹏 阅读量:
基于图神经网络的视频推荐系统 - 中兴通讯技术
您当前访问的的浏览器版本过低,为了给您带来更好的体验,建议您升级至Edge浏览器或者推荐使用Google浏览器
取消

基于图神经网络的视频推荐系统

作者:高宸,李勇,金德鹏 阅读量:1549

基于图神经网络的视频推荐系统

高宸,李勇,金德鹏
(清华大学,中国 北京 100084)

摘要:提出了一种基于图神经网络的视频推荐算法,将用户的视频观看序列型行为建模为图结构,用结点代表用户与视频,用边代表行为,引入两种类型的向量传播方法分别对用户的长期兴趣与短时兴趣进行建模。其中,通过用户结点与视频结点的双向传播刻画长期兴趣,借助视频结点切换关系的单向传播刻画短时兴趣,并通过多层向量传播实现对图上高阶邻接信息的捕捉。在一个真实世界的视频网站观看数据集上的实验表明,提出的方法与现有最佳方法相比,其推荐精准度得到了有效提升。进一步的实验表明,该方法能够有效缓解数据稀疏性的问题。 
关键词:视频推荐系统;用户兴趣建模;图神经网络;深度学习 


Video Recommender System with Graph Neural Networks

GAO Chen, LI Yong, JIN Depeng
(Tsinghua University, Beijing 100084, China)

Abstract: A novel recommendation model with graph neural networks is proposed. Users’ sequential video-watching behaviors are first constructed as a graph, which represents users and videos as nodes, and behaviors as edges. Then two kinds of embedding propagation methods are introduced for capturing users’ long-term and short-term preferences, respectively. Specifically, a user-item bi-directional embedding propagation layer is used for capturing long-term preferences while an item-item embedding propagation layer for capturing short-term preferences. Moreover, the multi-layer propagation is proposed to extract high-order connectivity. Experiments on a real-world video-watching dataset verify that the proposed method can outperform the state-of-the-art methods. Further experiments demonstrate that the proposed method can effectively alleviate the data sparsity issue.
Keywords: video recommender system; user preference modeling; graph neural network; deep learning

在线PDF浏览:PDF
PDF