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

In three linesComparative study of few-shot biomedical relation extraction with LLMs vs supervised learning on BioREDirect. Pairwise classification vs joint generation: F1=0.44 (few-shot) vs 0.56 (supervised) in micro-F1, but 0.45 vs 0.38 in macro-F1. LLMs outperform baseline on rare relations.

Summary generated by Claude — human-verified