Biomedical document triage using a hierarchical attention-based capsule network.

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Machine Learning

Background:

• Biomedical document triage is crucial for information extraction and precision medicine.
• Current neural network methods struggle with long, complex biomedical documents and capturing inter-sentence features.
• Effective classification of biomedical literature is essential for advancing medical research.

Purpose of the Study:

• To propose a novel hierarchical attention-based capsule model for biomedical document triage.
• To address the limitations of existing methods in handling long and complex biomedical texts.
• To improve the accuracy and efficiency of automatic biomedical document classification.

Main Methods:

• Development of a hierarchical attention mechanism to focus on salient information across sentences.
• Integration of capsule networks to effectively represent and aggregate document features.
• Evaluation of the proposed model on three public biomedical document corpora.

Main Results:

• The hierarchical attention mechanism effectively captures valuable features across sentences.
• Capsule networks contribute to constructing a robust latent feature representation for documents.
• The proposed model demonstrates strong performance in biomedical document triage.

Conclusions:

• Both hierarchical attention and capsule networks significantly benefit the biomedical document triage task.
• The developed model is competitive with and often superior to existing state-of-the-art methods.
• This work offers an effective solution for classifying complex biomedical documents, supporting precision medicine.