MIEM‐CA: A Multigranularity Information‐Enhanced Entity Matching Method Based on Collaborative Agents.
Saved in:
| Title: | MIEM‐CA: A Multigranularity Information‐Enhanced Entity Matching Method Based on Collaborative Agents. |
|---|---|
| Authors: | Xu, Yaoli1 (AUTHOR), Yang, Zheran1 (AUTHOR), Liang, Shuaixi2 (AUTHOR), Liu, Yongwen1 (AUTHOR), Xie, Chunli1 (AUTHOR), Zhai, Haojie1 (AUTHOR), Shi, Xiayang1 (AUTHOR) 2020040@zzuli.edu.cn, Murray, Richard (AUTHOR) rmurray@wiley.com |
| Source: | International Journal of Intelligent Systems. 6/19/2026, Vol. 2026, p1-30. 30p. |
| Subjects: | Multiagent systems, Encoding, Machine learning, Language models, Data integration, Information processing, Information storage & retrieval systems |
| Abstract: | Entity matching (EM) aims to identify records from different data sources referring to the same real‐world entity. Despite remarkable advances with pretrained language models (PLMs), existing PLM‐based matchers still encounter significant challenges in effectively integrating external knowledge, representing semantic information at multiple granularities, and handling numerical snippets. To address these challenges, we propose a multigranularity information‐enhanced EM method based on collaborative agents (MIEM‐CA), featuring three key components: (1) a multiagent information enhancement module (MI) that leverages extensive external knowledge, the decision‐making and collaboration capabilities of autonomous agents, and the semantic comprehension power of large language models (LLMs), by integrating attribute selection, web search, and feature extraction agents to improve the completeness of entity representation; (2) a multigranularity semantic encoder (ME) that incrementally captures and integrates token‐, attribute‐, and entity‐level semantics, along with their cross‐level correlations, across hierarchical representations spanning the token, attribute, and entity layers (ELs); and (3) a numerical‐aware agent module (NA) that employs the chain‐of‐thought (CoT) strategy to extract numerical information effectively, leverages LLMs to infer the semantic types of these numerical values, and calculates their semantic‐aware numerical similarity. Comprehensive experiments on 10 benchmark datasets, which cover structured, dirty, and textual EM settings, demonstrate that, compared with five baseline methods, MIEM‐CA achieves an average F1 score improvement of 6.35% on structured datasets, 9.07% on the dirty datasets, and 8.11% across all datasets. [ABSTRACT FROM AUTHOR] |
| Copyright of International Journal of Intelligent Systems is the property of Wiley-Blackwell and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.) | |
| Database: | Engineering Source |
|
Full text is not displayed to guests.
Login for full access.
|
|
| Abstract: | Entity matching (EM) aims to identify records from different data sources referring to the same real‐world entity. Despite remarkable advances with pretrained language models (PLMs), existing PLM‐based matchers still encounter significant challenges in effectively integrating external knowledge, representing semantic information at multiple granularities, and handling numerical snippets. To address these challenges, we propose a multigranularity information‐enhanced EM method based on collaborative agents (MIEM‐CA), featuring three key components: (1) a multiagent information enhancement module (MI) that leverages extensive external knowledge, the decision‐making and collaboration capabilities of autonomous agents, and the semantic comprehension power of large language models (LLMs), by integrating attribute selection, web search, and feature extraction agents to improve the completeness of entity representation; (2) a multigranularity semantic encoder (ME) that incrementally captures and integrates token‐, attribute‐, and entity‐level semantics, along with their cross‐level correlations, across hierarchical representations spanning the token, attribute, and entity layers (ELs); and (3) a numerical‐aware agent module (NA) that employs the chain‐of‐thought (CoT) strategy to extract numerical information effectively, leverages LLMs to infer the semantic types of these numerical values, and calculates their semantic‐aware numerical similarity. Comprehensive experiments on 10 benchmark datasets, which cover structured, dirty, and textual EM settings, demonstrate that, compared with five baseline methods, MIEM‐CA achieves an average F1 score improvement of 6.35% on structured datasets, 9.07% on the dirty datasets, and 8.11% across all datasets. [ABSTRACT FROM AUTHOR] |
|---|---|
| ISSN: | 08848173 |
| DOI: | 10.1155/int/8100559 |
