Scholarlink

<p>Blocking is a crucial subtask within the field of entity matching. Although deep learning approaches have gained popularity for this task, they still face two primary challenges: (1) inadequate utilization of contextual semantics, and (2) sensitivity to noise. To address these limitations, we introduce a novel framework, named <b>P</b>re-trained language model and <b>M</b>asked <b>A</b>uto<b>e</b>ncoder (PMAE). We leverage a pre-trained language model to extract rich semantic information. Furthermore, we incorporate noise into the semantic representations of entities through a dual-branch masking mechanism, to enhance the model’s robustness to noisy data. Through extensive evaluations against several state-of-the-art methods on benchmark datasets, PMAE demonstrates significant improvements, achieving an average enhancement of 4.45% in the recall, outperforming existing techniques. The code is avaible at <a href="https://github.com/SovereignLin/PMAE">https://github.com/SovereignLin/PMAE</a></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

Enhancing blocking for entity matching with pre-trained language model and masked autoencoder

  • 期刊论文

  • 发表日期：

    2026年7月24日

查看全文
  • Zelin Li,
  • Ke Li,
  • Shuo Wang

摘要

Blocking is a crucial subtask within the field of entity matching. Although deep learning approaches have gained popularity for this task, they still face two primary challenges: (1) inadequate utilization of contextual semantics, and (2) sensitivity to noise. To address these limitations, we introduce a novel framework, named Pre-trained language model and Masked Autoencoder (PMAE). We leverage a pre-trained language model to extract rich semantic information. Furthermore, we incorporate noise into the semantic representations of entities through a dual-branch masking mechanism, to enhance the model’s robustness to noisy data. Through extensive evaluations against several state-of-the-art methods on benchmark datasets, PMAE demonstrates significant improvements, achieving an average enhancement of 4.45% in the recall, outperforming existing techniques. The code is avaible at https://github.com/SovereignLin/PMAE