《计算机应用》唯一官方网站 ›› 2025, Vol. 45 ›› Issue (6): 1793-1800.DOI: 10.11772/j.issn.1001-9081.2024050703

• 人工智能 • 上一篇    

融合三元组和文本属性的多视图实体对齐

翟社平1,2, 黄妍1(), 杨晴1, 杨锐1   

  1. 1.西安邮电大学 计算机学院,西安 710721
    2.陕西省网络数据分析与智能处理重点实验室(西安邮电大学),西安 710121
  • 收稿日期:2024-05-30 修回日期:2024-07-30 接受日期:2024-08-28 发布日期:2024-09-09 出版日期:2025-06-10
  • 通讯作者: 黄妍
  • 作者简介:翟社平(1971—),男,陕西宝鸡人,教授,博士,CCF高级会员,主要研究方向:语义计算、区块链
    黄妍(2000—),女,陕西渭南人,硕士研究生,主要研究方向:知识图谱 muhy0402@163.com
    杨晴(2000—),女,陕西渭南人,硕士研究生,主要研究方向:知识图谱
    杨锐(1976—),女,陕西咸阳人,讲师,硕士,主要研究方向:知识图谱。
  • 基金资助:
    国家自然科学基金资助项目(61373116);工业和信息化部通信软科学项目(2017-R-22);陕西省重点研发计划项目(2022GY-038);陕西省教育厅科学研究计划项目(18JK0697);陕西省大学生创新创业训练计划项目(202211664053);西安邮电大学研究生创新基金资助项目(CXJJYL2022052)

Multi-view entity alignment combining triples and text attributes

Sheping ZHAI1,2, Yan HUANG1(), Qing YANG1, Rui YANG1   

  1. 1.School of Computer Science and Technology,Xi’an University of Posts and Telecommunications,Xi’an Shaanxi 710121,China
    2.Shaanxi Key Laboratory of Network Data Analysis and Intelligent Processing (Xi’an University of Posts and Telecommunications),Xi’an Shaanxi 710121,China
  • Received:2024-05-30 Revised:2024-07-30 Accepted:2024-08-28 Online:2024-09-09 Published:2025-06-10
  • Contact: Yan HUANG
  • About author:ZHAI Sheping, born in 1971, Ph. D., professor. His research interests include semantic computing, blockchain.
    HUANG Yan, born in 2000, M. S. candidate. Her research interests include knowledge graph.
    YANG Qing, born in 2000, M. S. candidate. Her research interests include knowledge graph.
    YANG Rui, born in 1976, M. S., lecturer. Her research interests include knowledge graph.
  • Supported by:
    National Natural Science Foundation of China(61373116);Communication Soft Science Project of Ministry of Industry and Information Technology(2017-R-22);Shaanxi Province Key Research and Development Program(2022GY-038);Scientific Research Program of Shaanxi Education Department(18JK0697);Innovation and Entrepreneurship Training Program for College Students in Shaanxi Province(202211664053);Xi’an University of Posts and Telecommunications Graduate Student Innovation Fund(CXJJYL2022052)

摘要:

实体对齐(EA)旨在识别不同来源的知识图谱(KG)中指代相同的实体。现有的EA模型大多关注实体自身的特征,部分模型引入了实体的关系和属性信息辅助实现对齐,然而这些模型忽视了实体中潜在的邻域信息和语义信息。为了解决上述问题,提出一种融合三元组和文本属性的多视图EA模型(MultiEA)。所提模型将实体信息分为多个视图以实现对齐。针对缺少邻域信息的问题,采用图卷积网络(GCN)与翻译模型来并行学习嵌入实体的关系信息;针对缺少语义信息的问题,采用词嵌入与预训练语言模型学习属性文本的语义信息。实验结果表明,在DBP15K的3个子数据集上,相较于得到最优结果的基线模型EPEA(Entity-Pair Embedding Approach for KG alignment),所提模型的Hits@1值分别提升了2.18、1.36和0.96个百分点,平均倒数排名(MRR)分别提升了2.4、0.9和0.5个百分点,验证了所提模型的有效性。

关键词: 实体对齐, 知识嵌入, 注意力机制, 依存句法分析, BERT

Abstract:

Entity Alignment (EA) is to identify entities referring to the same thing in the Knowledge Graphs (KGs) of different sources. Most of the existing EA models focus on characteristics of the entities themselves, some of the models introduce entity relationship and attribute information to assist in alignment. However, these models ignore potential neighborhood information and semantic information in the entities. In order to solve the above problems, a Multi-view EA model combining triples and text attributes (MultiEA) was proposed. In the proposed model, entity information was divided into multiple views to achieve alignment. For the lack of neighborhood information, Graph Convolutional Network (GCN) and translation model were used to learn relationship information embedded in entities in parallel. Aiming at the lack of semantic information, word embedding and pre-trained language model were adopted to learn semantic information of attribute text. Experimental results show that on the three sub-datasets of DBP15K, compared to the baseline model EPEA (Entity-Pair Embedding Approach for KG alignment) that yields the optimal results, the Hits@1 value of the proposed model is increased by 2.18,1.36 and 0.96 percentage points, respectively, and the Mean Reciprocal Rank (MRR) of the proposed model is improved by 2.4,0.9 and 0.5 percentage points, respectively, indicating the effectiveness of the proposed model.

Key words: Entity Alignment (EA), knowledge embedding, attention mechanism, dependency syntactic parsing, BERT (Bidirectional Encoder Representations from Transformers)

中图分类号: