Building an intelligent diabetes Q&A system with knowledge graphs and large language models.

Area of Science:

• Artificial Intelligence
• Medical Informatics
• Computational Linguistics

Background:

• Traditional healthcare systems struggle with complex medical queries for diabetes management.
• Personalized medical information delivery is crucial for effective diabetes care.
• Existing question-answering systems lack the contextual understanding for nuanced medical guidance.

Purpose of the Study:

• To introduce an intelligent question-answering system for personalized medical information delivery to diabetic patients.
• To enhance the accuracy and contextual relevance of medical guidance for diabetes management.
• To address the limitations of current healthcare systems in handling complex patient queries.

Main Methods:

• Integration of large language models (Baichuan2-13B, Qwen2.5-7B) with a Neo4j-based knowledge graph.
• Application of Low-Rank Adaptation (LoRA) and prompt-based learning techniques to improve semantic understanding and response generation.
• Evaluation of system performance through entity recognition and intent classification tasks.

Main Results:

• Achieved 85.91% precision in entity recognition and 88.55% precision in intent classification.
• Demonstrated significant improvement in accuracy and clinical relevance through knowledge graph integration.
• Showcased enhanced capability in providing personalized medical responses for diabetes management.

Conclusions:

• The integration of large language models and knowledge graphs effectively improves medical question-answering systems.
• The proposed approach offers a promising framework for advancing diabetes management through personalized healthcare.
• This methodology provides a foundation for future personalized interventions in various healthcare applications.