Scholarlink

<p>Entity Alignment (EA) is crucial for knowledge graph integration, as it focuses on matching equivalent entities from various knowledge graphs to establish a consolidated, integrated knowledge graph. Temporal Knowledge Graphs (TKGs) enhance static knowledge graphs by incorporating temporal information, garnering increasing interest in the research community. Recent studies on TKGs reveal that temporal information significantly boosts EA performance. However, existing TKG research methods often overlook the varying degrees of influence that neighbor information has on EA. We present a novel method, Temporal Neighborhood-aware Entity Alignment (TNEA), that effectively combines the structural, attribute, and temporal information of aligned entities from various temporal knowledge graphs (TKGs). Unlike existing methods, TNEA incorporates attention coefficients on relation and attribute features, enabling it to represent and comprehend the significance of various information sources contributing to entity alignment. To process attribute information and temporal information in TKG, we utilize the multilingual BERT model for attribute data encoding and the LSTM model for time-related data encoding. Concurrently, the Graph Attention Network is employed to assimilate the structural and attribute information from the TKGs. We carry out a series of tests on time-related datasets to confirm the effectiveness of our approach. The results indicate that TNEA surpasses current approaches.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Neighborhood-aware entity alignment for temporal knowledge graph

  • 期刊论文

  • 发表日期：

    2025年4月22日

查看全文
  • Lin Zhu,
  • Guishun Li,
  • Luyi Bai

摘要

Entity Alignment (EA) is crucial for knowledge graph integration, as it focuses on matching equivalent entities from various knowledge graphs to establish a consolidated, integrated knowledge graph. Temporal Knowledge Graphs (TKGs) enhance static knowledge graphs by incorporating temporal information, garnering increasing interest in the research community. Recent studies on TKGs reveal that temporal information significantly boosts EA performance. However, existing TKG research methods often overlook the varying degrees of influence that neighbor information has on EA. We present a novel method, Temporal Neighborhood-aware Entity Alignment (TNEA), that effectively combines the structural, attribute, and temporal information of aligned entities from various temporal knowledge graphs (TKGs). Unlike existing methods, TNEA incorporates attention coefficients on relation and attribute features, enabling it to represent and comprehend the significance of various information sources contributing to entity alignment. To process attribute information and temporal information in TKG, we utilize the multilingual BERT model for attribute data encoding and the LSTM model for time-related data encoding. Concurrently, the Graph Attention Network is employed to assimilate the structural and attribute information from the TKGs. We carry out a series of tests on time-related datasets to confirm the effectiveness of our approach. The results indicate that TNEA surpasses current approaches.