Optimized biomedical entity relation extraction method with data augmentation and classification using GPT-4 and...

Area of Science:

• Computational Biology
• Natural Language Processing (NLP)
• Bioinformatics

Background:

• Biomedical relation extraction remains a challenge despite extensive research.
• Existing natural language processing (NLP) techniques require further improvement for accuracy.
• Large language models (LLMs) present a novel opportunity to enhance NLP task performance.

Purpose of the Study:

• To improve the accuracy of biomedical relation extraction tasks.
• To leverage the capabilities of advanced large language models (LLMs) for enhanced NLP performance.
• To present a novel method integrating Gemini and GPT-4 for relation extraction.

Main Methods:

• Utilized GPT-4 for generating augmented training data.
• Employed an ensemble learning technique to combine diverse model outputs for precise predictions.
• Fine-tuned the BioNLP-PubMed-Bert classification model using Gemini responses as input.

Main Results:

• Achieved improved performance in precision, recall, and F1 scores on the BioCreative VIII Track 01 dataset.
• Demonstrated the efficacy of integrating Gemini and GPT-4 for biomedical relation extraction.
• The fine-tuned BioNLP-PubMed-Bert model showed significant performance gains.

Conclusions:

• The proposed method effectively enhances biomedical relation extraction accuracy.
• Integrating LLMs like Gemini and GPT-4 offers a promising direction for advancing NLP in bioinformatics.
• The ensemble learning approach combined with fine-tuning yields superior results for complex NLP tasks.