Knowledge-enhanced Graph Topic Transformer for Explainable Biomedical Text Summarization

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Artificial Intelligence

Background:

• The rapid growth of biomedical literature necessitates effective automatic summarization.
• Current pre-trained language models (PLMs) for summarization lack domain-specific knowledge, leading to incoherent and incomplete summaries.
• Explainability is critical for trust and understanding in biomedical text summarization.

Purpose of the Study:

• To develop a novel model for explainable biomedical text summarization.
• To improve the accuracy and coherence of summaries generated from biomedical literature.
• To address the limitations of existing PLM-based methods by incorporating domain knowledge.

Main Methods:

• Proposed a domain knowledge-enhanced graph topic transformer (DORIS) model.
• Integrated graph neural topic modeling with domain-specific knowledge from the Unified Medical Language System (UMLS).
• Fine-tuned transformer-based PLMs to enhance summarization capabilities.

Main Results:

• DORIS outperforms existing state-of-the-art PLM-based methods in biomedical extractive summarization.
• The model demonstrates improved accuracy and coherence in generated summaries.
• Graph neural topic modeling provides inherent explainability, clarifying sentence selection processes.

Conclusions:

• DORIS offers a significant advancement in explainable biomedical text summarization.
• Integrating domain knowledge and graph topic modeling enhances summary quality and transparency.
• The model provides a more understandable and reliable approach to summarizing complex biomedical information.