Computer Engineering and Applications ›› 2024, Vol. 60 ›› Issue (10): 173-179.DOI: 10.3778/j.issn.1002-8331.2305-0252

• Pattern Recognition and Artificial Intelligence • Previous Articles     Next Articles

Item Recommendation Algorithm Integrating Knowledge Graph and Attention Mechanism

XING Junye, XING Xing, JIA Zhichun, WANG Hongda, LIU Jiawen   

  1. School of Information Sciences and Technology, Bohai University, Jinzhou, Liaoning 121000, China
  • Online:2024-05-15 Published:2024-05-15

融合知识图谱与注意力机制的项目推荐算法

邢峻也,邢星,贾志淳,王鸿达,刘嘉雯   

  1. 渤海大学 信息科学与技术学院,辽宁 锦州 121000

Abstract: Knowledge graph contains rich semantic information, and is widely used in different recommendation scenarios. However, the existing recommendation method based on knowledge graph considers the relationship between user and project interaction in a coarse-grained way in the modeling process of graph neural network, and ignores the importance of different historical items when constructing user feature models. To solve such problems, the item recommendation algorithm that integrates knowledge graph and attention mechanism is proposed. The algorithm aggregates user characteristics and uses the attention mechanism to learn the high-order latent relationship of knowledge graph. When constructing item features, the algorithm propagates the item embedding representation between their neighborhoods, further integrates the graph convolutional network, and finally uses multi-layer neural networks for prediction. On the two sets of public datasets, the recall rate is improved by up to 6.9% by comparing the baseline algorithm.

Key words: recommendation system, knowledge graph, attention mechanism, graph convolutional network

摘要: 知识图谱蕴含丰富的语义信息,广泛应用在不同的推荐场景中。现有的基于知识图谱的推荐方法在图神经网络的建模过程中,粗粒度地考虑用户和项目交互的关系,构建用户特征模型时,忽略不同历史项目的重要性。针对此类问题,提出一种融合知识图谱与注意力机制的项目推荐算法。该算法聚合用户特征,使用注意力机制学习知识图谱高阶潜在关系,构建项目特征时传播其邻域之间的项目嵌入表示,用图卷积网络进行特征聚合,最后使用多层神经网络进行预测。该算法在两组公开数据集上,与基线算法进行对比实验,召回率最高提升6.9%。

关键词: 推荐系统, 知识图谱, 注意力机制, 图卷积神经网络