COTKEG: A Chain-of-Thought and Knowledge Graph Enhanced Method for Medical Large Language Models

With the rapid advancement of large language models (LLMs) in natural language processing (NLP), their medical applications are gaining attention. LLMs can support healthcare professionals in diagnosis and treatment but face challenges such as hallucinations, limited reasoning, and constrained data. High accuracy and reliability are crucial in medicine. This paper proposes the COTKEG framework, integrating chain-of-thought reasoning and knowledge graphs to enhance LLMs’ medical accuracy. The framework includes two key modules: one deconstructs complex medical issues step by step, while the other incorporates structured knowledge to reduce hallucinations and improve reliability. Experiments on MMCUMedical and CMB-Exam datasets show that COTKEG outperforms baselines in exact match (EM), partial correct rate (PCR), and medical expertise (h1) and harmlessness (h2) scores. Future work can optimize retrieval, adapt to dynamic knowledge, and expand applications. COTKEG offers a promising approach to improving medical LLMs’ accuracy, reliability, and safety.