Artificial Intelligence #few-shot learning#biomedical relation extraction
Few-Shot Biomedical Relation Extraction with LLMs: A Viable Alternative to Supervised Learning?
A new study on arXiv investigates few-shot biomedical relation extraction using large language models (LLMs). The best model achieved micro-F1 of 0.44, surpassing prior few-shot results but below supervised baseline. However, on macro-F1, prompt-based methods outperformed supervised learning, particularly on rare relation types, highlighting LLMs' potential in low-resource settings.
Jun 16, 2026 1 source