A Chinese question and answer system for liver cancer based on knowledge graph and large language mode

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Natural Language Processing

Background:

• Existing question-and-answer (Q&A) systems lack specialized liver cancer information and struggle with Chinese language queries.
• There is a need for a convenient and reliable method for patients to access liver cancer-related data.

Purpose of the Study:

• To develop a specialized Q&A system for liver cancer using knowledge graphs and Large Language Models (LLMs).
• To enhance the understanding of Chinese medical questions and provide clinically relevant answers.

Main Methods:

• A knowledge graph was constructed using data from clinical electronic medical records and the xywy.com medical website.
• ChatGLM3.5 and BERT were employed for entity extraction and user intent recognition, respectively.
• Information retrieval from the knowledge graph and natural language generation were used to formulate responses.

Main Results:

• The system achieved 92.34% precision in intent classification, outperforming BERT and GEBERT models.
• Responses demonstrated higher relevance and better alignment with patients' natural language patterns.
• The developed system significantly improved the usefulness and reliability of liver cancer information access.

Conclusions:

• The integration of knowledge graphs and LLMs offers a robust solution for specialized medical Q&A systems.
• This approach effectively addresses the limitations of existing systems in handling domain-specific queries and clinical data.
• The enhanced liver cancer Q&A system provides a valuable tool for patients seeking disease information.