Journal of Frontiers of Computer Science and Technology ›› 2021, Vol. 15 ›› Issue (11): 2048-2062.DOI: 10.3778/j.issn.1673-9418.2103086

• Surveys and Frontiers • Previous Articles     Next Articles

Knowledge Graph Embedding Technology: A Review

SHU Shitai, LI Song, HAO Xiaohong, ZHANG Liping   

  1. School of Computer Science and Technology, Harbin University of Science and Technology, Harbin 150080, China
  • Online:2021-11-01 Published:2021-11-09

知识图谱嵌入技术研究进展

舒世泰李松郝晓红张丽平   

  1. 哈尔滨理工大学 计算机科学与技术学院,哈尔滨 150080

Abstract:

Knowledge graph embedding (KGE) is a new research hotspot in the field of knowledge graphs, which aims to apply the translation invariance of word vectors to embedding entities and relationships of the knowledge graph into a low-dimensional vector space to complete knowledge representation. In this paper, it is mainly concerned with the classification according to the types of practical problems to be solved. Firstly, it expounds four major types of embedding methods of knowledge graph, including deep learning-based methods, graphical features-based methods, translation model-based methods, and other model-based methods. The algorithm ideas of each model are elaborated, and the advantages and disadvantages of each model are concluded. Secondly, the algorithm experi-ment of knowledge graph embedding is analyzed and summarized from the four aspects of commonly used data sets, evaluation indicators, algorithms, and experiments, then a horizontal and vertical comparison of the embedding method is made. Finally, from the perspective of solving practical problems, the future direction of knowledge graph embedding technology is given. Through research, it is discovered that in the deep learning-based method, LCPE achieves the best effect; in the graphical features-based method, TCE makes the best impression; whereas in the translation model-based method, NTransGH responds most optimistically. Future researches can be expanded on the basis of LCPE, TCE, and NTransGH to continuously improve the experimental effects of link prediction and triplets classification.

Key words: knowledge graph embedding (KGE), knowledge representation, knowledge graph completion (KGC), link prediction, triple classification

摘要:

知识图谱嵌入(KGE)是知识图谱领域一个新的研究热点,旨在利用词向量的平移不变性将知识图谱中实体和关系嵌入到低维向量空间,进而完成知识表示。以解决实际问题的类型为划分依据,首先,阐述了四类主要的知识图谱嵌入方法,包括基于深度学习的方法、基于图形特征的方法、基于翻译模型的方法以及基于其他模型的方法,对每种模型的算法思想进行详细阐述,总结了每种模型的优缺点;其次,从常用数据集、评价指标、算法、实验四方面对知识图谱嵌入算法实验进行分析与归纳,对嵌入方法做了横纵向对比;最后,从解决实际问题的角度出发,给出了知识图谱嵌入技术未来的发展方向。通过研究,发现在基于深度学习的方法中,LCPE模型的效果最好;在基于图形特征的方法中,TCE模型的效果最好;在基于翻译模型的方法中,NTransGH模型的效果最好。今后的研究可以在LCPE、TCE、NTransGH的基础上进行拓展,不断提高链接预测和三元组分类的实验效果。

关键词: 知识图谱嵌入(KGE), 知识表示, 知识图谱补全(KGC), 链接预测, 三元组分类