作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2018, Vol. 44 ›› Issue (7): 205-211. doi: 10.19678/j.issn.1000-3428.0047598

• 人工智能及识别技术 • 上一篇    下一篇

基于结构与文本联合表示的知识图谱补全方法

鲍开放,顾君忠,杨静   

  1. 华东师范大学 计算机科学技术系,上海 200062
  • 收稿日期:2017-06-14 出版日期:2018-07-15 发布日期:2018-07-15
  • 作者简介:鲍开放(1992—),男,硕士研究生,主研方向为自然语言处理;顾君忠,教授、博士生导师;杨静,副教授。
  • 基金资助:

    国家科技支撑计划项目(2015BAH01F02)。

Knowledge Graph Completion Method Based on Jointly Representation of Structure and Text

BAO Kaifang,GU Junzhong,YANG Jing   

  1. Department of Computer Science and Technology,East China Normal University,Shanghai 200062,China
  • Received:2017-06-14 Online:2018-07-15 Published:2018-07-15

摘要:

现有的表示学习算法不能很好地表示知识图谱中的复杂关系,且未能充分利用实体的描述文本。为此,建立一种结合文本表示和结构表示的联合表示学习模型。使用深度卷积神经网络对实体的描述文本进行编码得到文本表示,通过引入非对称映射操作的基于翻译思想的模型生成结构表示,将两者进行联合学习从而得到实体和关系表示,同时使用不同的低秩矩阵分别对头实体和尾实体进行映射,使其能更好地表现知识图谱中的复杂关系。实验结果表明,相对文本表示和结构表示的单独训练模型,该模型具有更好的表示性能。

关键词: 知识图谱补全, 表示学习, 深度学习, 词向量, 知识表示

Abstract:

The existing representation learning algorithms can not well represent the complex relationship in knowledge graph,and fails to make full use of the description text of entities.To solve this problem,this paper proposes a jointly representation learning model combining text representation and structure representation.The deep Convolution Neural Network(CNN) is used to encode the text of the entity to get the text representation,the structure representation is generated by introducing the translation thought model of asymmetric mapping operation,and the two are jointly studied to get entity and relation representation.At the same time,different low-rank matrices is used to project the head entity and the tail entity separately,so that the proposed model can better express the complex relationship in knowledge graph.Experimental results show that the proposed model has better representation ability than the single training model of text representation and structure representation.

Key words: knowledge graph completion, representation learning, deep learning, word vector, knowledge representation

中图分类号: