图书情报工作 ›› 2020, Vol. 64 ›› Issue (21): 99-110.DOI: 10.13266/j.issn.0252-3116.2020.21.013

• 知识组织 • 上一篇    下一篇

融合结构与文本特征的知识图谱关系预测方法研究

林泽斐1,2, 欧石燕1   

  1. 1. 南京大学信息管理学院 南京 210093;
    2. 福建师范大学社会发展学院 福州 350007
  • 收稿日期:2020-06-09 修回日期:2020-08-19 出版日期:2020-11-05 发布日期:2020-11-05
  • 通讯作者: 欧石燕(ORCID:0000-0001-8617-6987),教授,博士生导师,通讯作者,E-mail:oushiyan@nju.edu.cn
  • 作者简介:林泽斐(ORCID:0000-0001-8637-7359),讲师,博士研究生。
  • 基金资助:
    本文系国家社会科学基金重点项目"基于关联数据的学术文献内容语义发布及其应用研究"(项目编号:17ATQ001)研究成果之一。

Research on Relation Prediction in Knowledge Graphs by Fusing Structure and Text Features

Lin Zefei1,2, Ou Shiyan1   

  1. 1 School of Information Management, Nanjing University, Nanjing 210093;
    2 College of Social Development, Fujian Normal University, Fuzhou 350007
  • Received:2020-06-09 Revised:2020-08-19 Online:2020-11-05 Published:2020-11-05
  • Supported by:
     

摘要: [目的/意义] 提出一种融合内部结构特征和外部文本特征的知识图谱关系预测新方法,旨在预测知识图谱中两实体间缺失关系的类型。[方法/过程] 将关系路径和反映实体间关系的文本矩阵化,通过卷积神经网络学习与特定关系类型相关的结构和文本模式特征,在此基础上训练模型实现关系预测。[结果/结论] 实验结果显示,该方法在评测数据集上的性能表现超过对照方法的水平,可有效提升知识图谱关系预测的性能。通过实际应用发现,该方法在知识服务中具有良好的应用价值。

 

关键词: 知识图谱, 关系预测, 特征融合, 深度学习

Abstract: [Purpose/significance] Relation prediction is an important task in knowledge graph completion, and plays an important role in improving the completeness of knowledge in knowledge graphs. The paper proposes a new relation prediction method that combines internal structure features and external text features, which aims to predict the missing relations between two entities in knowledge graphs. [Method/process] The method transforms the relation paths in a knowledge graph and the texts that involve entity relationships into matrixes, learns the structure features and text pattern features related to a specific relation type through convolutional neural networks, and then trains a model based on the learned features for relation prediction. [Result/conclusion] The results shows that the performance of our proposed method on evaluation data sets is superior to the state-of-the-art approaches, and the method can effectively improves the performance of knowledge graph relationship prediction. Through practical application, it wvas found that this method has high application value in knowledge services.

Key words: knowledge graph, relation prediction, feature fusion, deep learning

中图分类号: