计算机工程与应用2025，Vol.61Issue(10)：50-65,16.DOI:10.3778/j.issn.1002-8331.2408-0350
基于图神经网络的知识推理方法研究综述
Review of Knowledge Reasoning Methods Based on Graph Neural Networks
摘要
Abstract
Knowledge reasoning is a fundamental task in knowledge graph completion.It has received widespread atten-tion from the academic community.With the development of knowledge reasoning technology,applying graph neural net-works to knowledge reasoning methods can fully consider the structural information of knowledge graphs,making them more interpretable and stronger in reasoning ability,which is currently one of the research hotspots.The basic concepts of knowledge graph and knowledge reasoning are described.Knowledge reasoning methods based on graph neural networks are categorized from the perspectives of closed-world and open-world settings.In the closed-world context,the emphasis is placed on two types of methods:graph convolutional networks and graph attention networks.In the open-world context,semi-inductive and fully-inductive methods are explored.The typical model frameworks of these methods are compared and analyzed,and their respective strengths and weaknesses are summarized.Finally,the applications of graph neural net-work reasoning in intelligent question answering,recommendation system and biomedicine are discussed,and the future research direction of graph neural network based knowledge reasoning is prospected.关键词
知识图谱/图神经网络/知识推理/封闭世界/开放世界/归纳推理Key words
knowledge graph/graph neural networks/knowledge reasoning/closed world/open world/inductive reasoning分类
计算机与自动化引用本文复制引用
刘雪洋,李卫军,刘世侠,丁建平,苏易礌..基于图神经网络的知识推理方法研究综述[J].计算机工程与应用,2025,61(10):50-65,16.基金项目
国家自然科学基金(62066038,61962001) （62066038,61962001）
宁夏自然科学基金(2021AAC03215) （2021AAC03215）
中央高校科研项目(2022PT_S04,2021JCYJ12) （2022PT_S04,2021JCYJ12）
北方民族大学研究生创新项目(YCX24127). （YCX24127）
