Related Experiment Video
Updated: May 24, 2026
07:35
A Knowledge Graph Approach to Elucidate the Role of Organellar Pathways in Disease via Biomedical Reports
Published on: October 13, 2023
Fine-Grained Mention-Level Analysis of Biomedical Entity Linking Models
1Université Paris-Saclay, CNRS, Laboratoire Interdisciplinaire des Sciences du Numérique, 91400, Orsay, France.
Studies in Health Technology and Informatics
|May 23, 2026
Summary
We developed a fine-grained evaluation for Biomedical Entity Linking (BEL) to reveal model weaknesses. This approach uncovers how mention characteristics impact performance, offering deeper insights than aggregate scores.
Area of Science:
- Computational Biology
- Natural Language Processing
- Bioinformatics
Background:
- Biomedical Entity Linking (BEL) is crucial for organizing biomedical text data.
- Current global evaluation metrics for BEL systems can mask specific model limitations.
- Understanding these limitations is key to improving automated knowledge extraction.
Purpose of the Study:
- To introduce a fine-grained evaluation framework for BEL systems.
- To analyze BEL model performance based on mention-level characteristics.
- To identify systematic weaknesses in both neural and rule-based BEL systems.
Main Methods:
- Developed a novel evaluation framework focusing on mention-level attributes.
- Analyzed performance across characteristics like mention length, lexical variation, synonymy, homonymy, and training frequency.
- Applied the framework to assess neural and rule-based BEL systems using the BELB benchmark.
Main Results:
- Performance degradation in BEL systems is linked to specific mention-level difficulties.
- Consistent performance drops were observed across characteristics related to limited training data coverage.
- The fine-grained analysis revealed model weaknesses not apparent in aggregate evaluation scores.
Conclusions:
- A fine-grained approach provides a more comprehensive understanding of BEL system performance.
- Mention-level characteristics are critical determinants of BEL accuracy.
- This framework aids in identifying and addressing systematic weaknesses in BEL models for better knowledge structuring.
Related Concept Videos
00:49Ligand Binding and Linkage
Allosteric proteins have more than one ligand binding site; the binding of a ligand to any of these sites influences the binding of ligands to the other sites. When a protein is allosteric, its binding sites are called coupled or linked. In the case of enzymes, the site that binds to the substrate is known as the active site and the other site is known as the regulatory site. When a ligand binds to the regulatory site, this leads to conformational changes in the protein that can influence the...
00:49Ligand Binding and Linkage
Allosteric proteins have more than one ligand binding site; the binding of a ligand to any of these sites influences the binding of ligands to the other sites. When a protein is allosteric, its binding sites are called coupled or linked. In the case of enzymes, the site that binds to the substrate is known as the active site and the other site is known as the regulatory site. When a ligand binds to the regulatory site, this leads to conformational changes in the protein that can influence the...