摘要
为解决推荐模型中的用户信息缺失和用户动态偏好问题,满足用户个性化需求,提出一种融合社交关系和时序特征的图神经网络推荐模型。该模型先构建社交关系图,并通过注意力机制得到基于社交关系的用户潜在特征;再构建用户-项目交互图,利用门控循环单元和注意力机制捕获交互信息,分别获得用户的时序特征和项目特征;最后将用户潜在特征与时序特征融合得到新的用户特征,并将其与项目特征进行融合,经过多层感知机得到最终推荐结果。在不同数据集上进行实验,结果表明,该模型能更好地处理用户信息缺失和用户动态偏好问题,进而提升推荐性能。相较于经典的图神经网络推荐模型,该模型的精确率和归一化折损累计增益比分别提高了4.0%和4.1%。
To address user information loss and user dynamic preference in the recommendation model to meet the individual user needs,a graph neural network recommendation model fusing social relationship and temporal features is proposed.The model first constructs a social relationship graph and obtains the potential features of users based on social relationships through an attention machine.Then,an user-project interaction graph is constructed,and the interaction information captured by gated recurrent unit and attention mechanism to obtain users’temporal features and project characteristics respectively.Finally,the user’s potential features and temporal features are fused to obtain new user features,which are then fused with the project features and the final recommendation results are obtained through the multi-layer perceptron.The experimental results on different data sets show that the model can better deal with user information loss and user dynamic preference so that the recommendation performance is improved.Compared with the classical graph neural network recommendation model,the accuracy and normalized cumulative gain ratio of the model under discussion are increased by 4.0%and 4.1%,respectively.
作者
胡胜利
王柳
HU Shengli;WANG Liu(School of Computer Science and Engineering,Anhui University of Science and Technology,Huainan 232001,China)
出处
《厦门理工学院学报》
2024年第5期51-59,共9页
Journal of Xiamen University of Technology
基金
安徽理工大学2022年研究生创新基金项目“基于图神经网络的GNN-ULS个性化推荐研究”(2022CX2121)。
关键词
推荐模型
图神经网络
社交关系
时序特征
注意力机制
门控循环单元
recommendation model
graph neural network
social relationships
temporal features
attention mechanism
gated recurrent unit
