Few-Shot Biomedical Relation Extraction with Large Language Models: A Viable Alternative to Supervised Learning?
Researchers explored the use of large language models (LLMs) for few-shot biomedical relation extraction, a task crucial for structuring knowledge from biomedical literature. Their experiments on the BioREDirect dataset compared two LLM task formulations: pairwise classification and joint generation. While pairwise classification offered higher recall, joint generation proved more precise and efficient. The best LLM approach achieved a micro-F1 score of 0.44, surpassing prior few-shot results and demonstrating strong performance on rare relation types, though it fell short of fully supervised methods. AI