Exploring Biomedical Relation Extraction through ChatGPT Augmentation and Dual Training

doi:10.5281/zenodo.10118014

Published November 12, 2023 | Version v1

Conference proceeding Open

Exploring Biomedical Relation Extraction through ChatGPT Augmentation and Dual Training

1. Miin Wu School of Computing, National Cheng Kung University, Tainan, Taiwan
2. Dept. of Statistics, National Cheng Kung University, Tainan, Taiwan
3. Cross College Elite Program, National Cheng Kung University, Tainan, Taiwan
4. Dept. of Electrical Engineering, National Cheng Kung University, Tainan, Taiwan

Abstract

Relation extraction in biomedical text mining faces challenges due to complex terminology and rapidly growing literature. Our research focuses on improving relation extraction through data augmentation and targeted dual training. We fine-tuned a PubMedBERT model and enriched it with GPT-4 generated examples, iteratively refining the process. A dual training approach focusing on chemical entities significantly improved F1 scores by 2.04% on average. Our strategies demonstrate the effectiveness of ChatGPT-based augmentation and selective dual training for advancing biomedical text mining.

This article is part of the Proceedings of the BioCreative VIII Challenge and Workshop: Curation and Evaluation in the era of Generative Models.

Files