Abstract: Knowledge reasoning, as an important part of knowledge graph, has always been a hot topic of research. With the continuous development of deep learning, the combination of various deep learning models and knowledge reasoning has attracted great attention and been warmly welcomed by a large number of scholars at home and abroad. In order to improve the accuracy of inferring new knowledge from existing knowledge, the combination of the two has been widely studied. Knowledge reasoning based on deep learning can dig deeper, more carefully and more accurately, effectively improve the utilization rate of entities, relations, attributes and text information in the rich knowledge base, and make the reasoning effect better. This paper briefly introduces the concept of knowledge graph and knowledge completion, focuses on the concept and basic principle of knowledge reasoning, and expands from three directions of knowledge representation learning, knowledge acquisition and knowledge computing application. The paper reviews the latest research progress of CTransR, PTransE, TKRL, HAAT, AMNRE, CLSP, HDSA and SDLM models based on deep learning. The unsolved problems, research directions and future development prospects of deep learning-based knowledge reasoning in theory, algorithm and application are summarized.

Key words: knowledge graph, knowledge completion, knowledge reasoning, deep learning

摘要： 知识推理作为知识图谱的重要一环，一直处于重点研究热门对象之中。随着深度学习的不断发展，多种深度学习模型与知识推理的结合引起了很大的重视，得到了大量国内外学者的热捧。为了提高从已有知识中推理出新知识的正确率，二者的结合被广泛研究。基于深度学习的知识推理可以挖掘得更深、更仔细、更精确，有效提高了丰富知识库中的实体、关系、属性和文本信息等的利用率，使推理效果更佳。通过简单介绍知识图谱以及知识补全概念，重点叙述知识推理的概念及基本原理，从知识表示学习、知识获取和知识计算应用三个方向展开，综述了基于深度学习的知识推理CTransR、PTransE、TKRL、HAAT、AMNRE、CLSP、HDSA和SDLM模型的最新研究进展；总结了基于深度学习的知识推理在理论、算法和应用方面尚未克服的问题、研究方向和未来发展前景。

关键词: 知识图谱, 知识补全, 知识推理, 深度学习