计算机工程与应用 ›› 2022, Vol. 58 ›› Issue (20): 141-147.DOI: 10.3778/j.issn.1002-8331.2103-0248

• 模式识别与人工智能 • 上一篇    下一篇

基于双端知识图的图注意推荐模型

孙伟,陈平华,熊建斌,申建芳   

  1. 1.广东工业大学 计算机学院,广州 510006
    2.广东技术师范大学 自动化学院,广州 510665
  • 出版日期:2022-10-15 发布日期:2022-10-15

Graph Attention Recommendation Model Based on Dual-End Knowledge Graph

SUN Wei, CHEN Pinghua, XIONG Jianbin, SHEN Jianfang   

  1. 1.School of Computer, Guangdong University of Technology, Guangzhou 510006, China
    2.School of Automation, Guangdong Polytechnic Normal University, Guangzhou 510665, China
  • Online:2022-10-15 Published:2022-10-15

摘要: 针对现有图神经网络在捕获知识图谱信息并进一步用于推荐时,侧重于项目端建模所存在的问题,提出一种基于双端知识图的图注意推荐模型。该模型通过从用户端和项目端在知识图谱上挖掘相关属性来有效增强推荐。从用户端角度,通过知识图谱中实体之间的联系传播用户兴趣,沿着知识图谱中用户的历史点击项来扩展用户的潜在兴趣;从项目端角度,通过捕获知识图谱中的高阶结构和语义信息,对每个实体的邻居抽样作为接收场,通过图注意获得实体-实体交互信息,以此建模高阶邻域信息,最后使用交叉熵损失函数进行训练。结果表明,所提模型在关于电影、书籍和音乐推荐的三个数据集上,有效提高了推荐的准确性和可解释性。

关键词: 推荐系统, 知识图谱, 图注意机制, 高阶连通性, 偏好传播

Abstract: In view of the existing graph neural network(GNN) in capturing knowledge graph(KG) information and further using it for recommendation, focusing on the problems of project-side modeling, this paper proposes graph attention recommendation model based on dual-end knowledge graph. This model effectively enhances recommendation by mining relevant attributes on KG from the user side and the item side. From the perspective of the user side, it spreads user interest through the connections between entities in the knowledge graph and expands the users potential interest along the users historical clicks in KG. From the project side perspective, by capturing high-level structure and semantic information in KG, it samples the neighbors of each entity as the receiving field. The method obtains entity-entity interaction information through graph attention, and models high-order neighborhood information, finally uses the cross-entropy loss function for training. The results show that the model proposed in this paper is superior to other advanced approaches in accuracy and reliability on the three data sets about movie, book and music recommendation.

Key words: recommender systems, knowledge graph, graph attention mechanism, higher-order connectivity, embedding propagation