摘要
针对知识图注意力网络(KnowledgeGraphAttentionNetwork,KGAT)推荐模型在整个知识图谱上传播信息,容易引入噪声的问题,提出一种改进的KGAT模型,通过将注意力嵌入传播层替换为注意力机制嵌入和信息过滤两个部分实现注意力得分机制,优化节点嵌入。在两个公共数据集Amazon-Book和Last-FM上分别进行对比实验,实验结果表明改进模型在recall和ndcg两项评价指标上都有提升,其中在Amazon-Book上分别提升了1.54%和1.68%,在Last-FM上分别提升了1.03%和1.96%,有效地改善了推荐结果。
In order to solve the problem that the knowledge graph attention network for recommendation spread information across the entire knowledge graph and are prone to introducing irrelevant entities, an improved knowledge graph attention network model is proposed. It implements the attention scoring mechanism by replacing the attentive embedding propagation layers with the attention mechanism embedding and information filtering to optimize node embedding. Comparative experiments were carried out on two public datasets Amazon-Book and Last-FM to compare with the best baseline model KGAT. The experiments show that the model improves recall and ndcg by 1.54% and 1.68% on Amazon-Book, and 1.03% and 1.96% on Last-FM. It can be seen that the improved model can effectively improve the results of recommendation.
作者
王志寅
Wang Zhiyin(Department of Computer Science and Technology,Taiyuan Normal University,Jinzhong 030600)
出处
《现代计算机》
2022年第18期36-41,共6页
Modern Computer
基金
行为数据分析技术研究及应用(XTCXZX-2018-002)。
关键词
推荐系统
知识图谱
图神经网络
注意力机制
recommended system
knowledge graph
graph neural network
attention mechanism
