Integrating graph convolutional networks to enhance prompt learning for biomedical relation extraction.

Area of Science:

• Natural Language Processing
• Bioinformatics
• Computational Linguistics

Background:

• Biomedical relation extraction identifies relationships between entities in medical texts.
• Current models often struggle with grammatical dependencies, limiting their effectiveness.
• Graph Convolutional Networks (GCNs) show promise in processing syntactic information.

Purpose of the Study:

• To develop an improved biomedical relation extraction model.
• To address limitations in handling syntactic dependencies in existing models.
• To enhance performance by integrating GCNs with prompt learning.

Main Methods:

• A novel model fusing GCNs-enhanced prompt learning for relation extraction.
• Integration of syntactic dependency information analyzed by GCNs into a prompt learning framework.
• Predicting [MASK] token labels for relation extraction based on integrated syntactic information.

Main Results:

• The proposed model achieved high F1 scores across multiple biomedical datasets: GAD (85.57%), ChemProt (80.15%), PGR (95.10%), and DDI (84.11%).
• Performance surpassed existing baseline models on these datasets.
• Demonstrated the effectiveness of integrating GCNs for syntactic dependency analysis.

Conclusions:

• Enhancing prompt learning with GCNs effectively incorporates syntactic information into biomedical relation extraction.
• The proposed method achieves excellent performance in biomedical relation extraction tasks.
• The integration of GCNs offers a promising direction for future relation extraction research.