基于联合知识表示学习的多模态实体对齐
作者:
作者单位:

(河北科技大学信息科学与工程学院,石家庄050018)

作者简介:

通讯作者:

E-mail: zxmhebust@163.com.

中图分类号:

TP391

基金项目:

国家自然科学基金项目(51271033);河北省自然科学基金项目(F2018208116);河北省研究生创新项目(CXZXZSS2018092).


Multi-modal entity alignment based on joint knowledge representation learning
Author:
Affiliation:

(College of Information Science and Engineering,Heibei University of Science and Technology,Shijiazhuang050018,China)

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    基于知识表示学习的实体对齐方法是将多个知识图谱嵌入到低维语义空间,通过计算实体向量之间的相似度实现对齐.现有方法往往关注文本信息而忽视图像信息,导致图像中实体特征信息未得到有效利用.对此,提出一种基于联合知识表示学习的多模态实体对齐方法(ITMEA).该方法联合多模态(图像、文本)数据,采用TransE与TransD相结合的知识表示学习模型,使多模态数据能够嵌入到统一低维语义空间.在低维语义空间中迭代地学习已对齐多模态实体之间的关系,从而实现多模态数据的实体对齐.实验结果表明,ITMEA在WN18-IMG数据集中能够较好地实现多模态实体对齐.

    Abstract:

    Knowledge representation learning(KRL) based entity alignment represents multi-knowledge graphs(KGs) as low-dimensional embeddings and realizes entity alignment by measuring the similarities between entity vectors. Existing approaches, however, frequently primarily focus on the text information but ignore the image information, and lead entity feature information of the image to being underutilized. To address this problem, we propose an approach for image-text multi-modal entity alignment(ITMEA) via joint knowledge representation learning. The approach joints multi-modal(image and text) data, and embeds multi-model data into a uniform low-dimensional semantic space by a knowledge representation learning model which combines the translating embedding model(TransE) with the dynamic mapping matrix embedding model(TransD). In low-dimensional semantic space, the link-mapping relations among the aligned multi-modal entities can be learned iteratively from the seed set. The learned link-mapping relations can be applied to unaligned entities. In this way, the multi-modal entity alignment can be implemented. Experiment results on WN18-IMG datasets show that the ITMEA can achieve the multi-modal entity alignment, and it is useful for the construction and completion of the open knowledge graph.

    参考文献
    相似文献
    引证文献
引用本文

王会勇,论兵,张晓明,等.基于联合知识表示学习的多模态实体对齐[J].控制与决策,2020,35(12):2855-2864

复制
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2020-12-02
  • 出版日期: 2020-12-20