A neural translating general hyperplane for knowledge graph embedding

作     者：Zhu, Qiannan Zhou, Xiaofei Zhang, Peng Shi, Yong 

作者机构：Chinese Acad Sci Inst Informat Engn Beijing Peoples R China Univ Chinese Acad Sci Sch Cyber Secur Beijing Peoples R China Ant Financial Serv Grp Hangzhou Zhejiang Peoples R China CAS Res Ctr Fictitious Econ & Data Sci Beijing Peoples R China Chinese Acad Sci Key Lab Big Data Min & Knowledge Management Beijing Peoples R China Univ Nebraska Coll Informat Sci & Technol Omaha NE 68182 USA 

出 版 物：《JOURNAL OF COMPUTATIONAL SCIENCE》 (计算科学杂志)

年 卷 期：2019年第30卷第Jan.期

页      面：108-117页

核心收录：

学科分类：08[工学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：National Key RD Program [2017YFB0803003] National Natural Science Foundation of China 

主　　题：Knowledge representation Knowledge embedding Knowledge graph completion Link prediction Neural network 

摘      要：For completing knowledge graph, many translation-based models, like that Trans(E and H) which embed a knowledge graph into a continuous vector space and encode relations as translation operations in that space, have achieved better performance. However, most of them have limitations in expressing complex relations for knowledge graph. In this paper, we propose a translation-neural based method NTransGH for knowledge graph completion. NTransGH combines translation mechanism for modeling relations as translation operations by generalized hyperplanes, and a neural network for capturing more complex interactions between entities and relations. We conduct experiment on two tasks link prediction and triplet classification with two datasets. Experimental results show that NTransGH has strong expression in mapping properties of complex relations, and achieves significant and consistent improvements over state-of-the-art embedding methods. This paper is an extension of our previous works [1]. (C) 2018 Elsevier B.V. All rights reserved.