Researchers have developed LongBEL, a novel framework for biomedical entity linking that leverages full-document context and a memory of previous predictions. Unlike existing systems that process mentions independently, LongBEL aims to improve consistency by considering document-level dependencies. The system is trained using cross-validated predictions to mitigate errors that can arise from cascading issues during inference. Experiments across five biomedical benchmarks in multiple languages demonstrate LongBEL's effectiveness, particularly in datasets where concepts frequently reappear within a single document. AI
IMPACT This new framework could improve the accuracy and consistency of information extraction from biomedical literature.
RANK_REASON The item is an academic paper detailing a new method for biomedical entity linking. [lever_c_demoted from research: ic=1 ai=1.0]
- Adam Remaki
- alphaXiv
- arXivLabs
- Bibliographic Explorer
- CatalyzeX Code Finder for Papers
- Connected Papers
- CORE Recommender
- DagsHub
- Gotit.pub
- Hugging Face
- Influence Flower
- Litmaps
- LongBEL
- ScienceCast
- scite Smart Citations
- SNOMED CT
- Unified Medical Language System
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →