SSGU-CD: A combined semantic and structural information graph U-shaped network for document-level Chemical-Disease...

Area of Science:

• Biomedical Natural Language Processing
• Information Extraction
• Computational Biology

Background:

• Document-level relation extraction is crucial for understanding chemical-disease interactions in biomedical texts.
• Existing sentence-level methods fail to capture comprehensive inter-sentence relationships.
• Biomedical information extraction requires methods that leverage document structure.

Purpose of the Study:

• To develop an advanced method for document-level Chemical-Disease (CD) interaction extraction.
• To address the limitations of sentence-level approaches in capturing holistic document information.
• To enhance the practical application of biomedical text information extraction.

Main Methods:

• Proposed SSGU-CD, a Semantic and Structural information Graph U-shaped network.
• Represented document semantic and structural information as graphs.
• Fused original document context information within the graph framework.
• Utilized a balanced cross-entropy loss function for model optimization.

Main Results:

• SSGU-CD demonstrated significant improvements in Chemical-Disease interaction extraction performance.
• Evaluated on document-level datasets CDR and BioRED, showing superior results.
• The framework effectively captured and fused semantic and structural information.

Conclusions:

• SSGU-CD offers a robust solution for document-level Chemical-Disease interaction extraction.
• The graph-based approach enhances the extraction of complex biomedical relationships.
• This method advances the field of biomedical text mining and information retrieval.