Abstract
INTRODUCTION: The liver cancer question-and-answer (Q&A) system is primarily intended to help patients access disease-related information more conveniently. However, there is currently no Q&A system specifically developed for liver cancer. Additionally, most existing Q&A systems lack real clinical data and have limited capability in understanding Chinese questions. METHODS: This paper proposes a Chinese liver cancer question-answering system based on knowledge graphs and Large Language Models (LLMs). To unify information from diverse sources, the system employs a knowledge graph to store entities and inter-entity relationships extracted from patients' clinical electronic medical records and the professional medical website xywy.com, which serves as the foundation for the system's responses. Specifically, ChatGLM3.5 is utilized to extract entity information from questions, while BERT is applied to understand users' intent. Subsequently, the system retrieves corresponding information from the knowledge graph. Finally, the retrieved information is integrated, and a natural language response is generated as the answer to the question. RESULTS: The experimental results indicate that in terms of intent classification, our system achieves a precision of 92.34%, representing an improvement of 1.38% over the BERT model and 4.32% over the GEBERT model. In terms of response relevance, the system's outputs are more aligned with patients' daily speech patterns and exhibit higher relevance to the target questions. DISCUSSION: In conclusion, the improved method significantly enhances the usefulness and reliability of the liver cancer Q&A system.