SynCABEL: Synthetic Contextualized Augmentation for Biomedical Entity Linking

In three linesSynCABEL uses LLMs to generate contextualized synthetic training data to address the scarcity of expert annotations in biomedical entity linking. The framework achieves SOTA results on MedMentions (English), QUAERO (French), and SPACCC (Spanish), reaching full human supervision performance with 60% less annotated data.

Summary generated by Claude — human-verified