What Do Biomedical NER and Entity Linking Benchmarks Measure? A Corpus-Centric Diagnostic Framework
Two new research papers explore the challenges and potential solutions for biomedical entity linking (BEL) and named entity recognition (NER). One paper introduces BeLink, a system that uses instruction-tuned generative models to improve the efficiency and accuracy of BEL, showing significant gains in linking accuracy and reduced inference time. The second paper presents a diagnostic framework to better understand what existing biomedical NER and EL benchmarks actually measure, highlighting substantial differences in corpus properties that affect evaluation signals and generalization demands. AI
IMPACT These papers offer advancements in biomedical NLP, potentially improving the efficiency and interpretability of tools used in medical research and applications.