Prompt Tuning in Biomedical Relation Extraction.

Area of Science:

• Biomedical Informatics
• Natural Language Processing

Background:

• Biomedical relation extraction (RE) is crucial for knowledge graphs and text mining.
• Adapting pre-trained language models (PLMs) for RE, particularly in low-data scenarios, remains a challenge.

Purpose of the Study:

• To explore prompt tuning for biomedical RE, focusing on its effectiveness in few-shot learning.
• To develop a simple, efficient, and high-performing prompt tuning model for biomedical RE.

Main Methods:

• Developed a customized prompt tuning model for biomedical RE.
• Evaluated the model on the chemical-protein relation (CHEMPROT) and drug-drug interaction (DDI) datasets.
• Compared performance against conventional fine-tuned PLMs in few-shot settings.

Main Results:

• The prompt tuning model outperformed conventional fine-tuned PLMs on both CHEMPROT and DDI datasets, including few-shot scenarios.
• Demonstrated superior performance without requiring external knowledge or additional computational resources.
• Achieved state-of-the-art results with a simple and efficient approach.

Conclusions:

• Prompt tuning is an effective strategy for enhancing biomedical RE capabilities of PLMs.
• The proposed model offers a robust and efficient solution for extracting complex relations from biomedical texts.
• This work advances the field by providing a powerful tool for biomedical text mining.