Few-Shot Biomedical Relation Extraction with Large Language Models: A Viable Alternative to Supervised Learning?

The study investigates few-shot biomedical relation extraction (BioRE) using large language models (LLMs) through prompt-based learning, comparing pairwise classification and joint generation task formulations. Experiments on the BioREDirect dataset show that the best-performing model achieves a micro-F1 score of 0.44, outperforming previous few-shot results and demonstrating better performance in macro-F1 metrics for rare relation types. This research highlights the viability of LLMs in low-resource biomedical settings, emphasizing the need for clear relation definitions to improve extraction accuracy.