Research on Disease Drug Knowledge Graph Based on Different Source Medical Guidelines and Large Language Model

In this study, we propose a method for constructing a medical knowledge graph by integrating diverse data sources and employing natural language processing techniques to enhance drug treatment information and clinical decision support. We utilized authoritative medical guidelines from the NHS and Mayo Clinic, applied the SciBERT model for drug entity recognition, and standardized drug names through the UMLS system. Additionally, the ERNIE Bot large language model was used to generate and verify drug side effect information, enriching the knowledge graph with side effect data. This approach resulted in a multi-dimensional knowledge graph that captures relationships between diseases, drugs, drug categories, and side effects. Compared to traditional methods, our approach offers more comprehensive data sources, improved entity recognition accuracy, enriched graph content, and enhanced decision support capabilities. Ultimately, this knowledge graph aids in the rapid and precise development of personalized treatment plans and minimizes drug side effect risks in complex clinical scenarios.