现代情报2025，Vol.45Issue(12)：63-76,14.DOI:10.3969/j.issn.1008-0821.2025.12.006
从患者生成健康数据到用药建议生成
From Patient-Generated Health Data to Medication Recommendations
摘要
Abstract
[Purpose/Significance]The rapid growth of patient-generated health data(PGHD)has provided a wealth of resources for personalized healthcare services.However,its informational potential remains largely untapped.This study aims to develop a patient medication information question answering(Q&A)model enhanced by graph retrieval-augmented generation.The model is designed to extract relevant information from PGHD and generate natural language Q&A.[Method/Process]To address this problem,the study constructed an intelligent medical Q&A model based on a Graph Retrieval-Augmented Generation(GraphRAG)model that integrates patient medication knowledge graphs with large language models.The framework comprised two key modules:(1)Knowledge Graph Module:the study designed diverse prompt templates to guide LLMs in automatically extracting medication relationships from PGHD and conducted a compara-tive analysis of different prompting strategies.Using the optimal extraction results,the study constructed a medication information knowledge graph via Neo4j.(2)Retrieval-Augmented Generation Module:Leveraging the LangChain frame-work,the study integrated LLMs with the knowledge graph to achieve GraphRAG,enabling the LLM to retrieve relevant facts from the knowledge graph and generate natural language answers.[Result/Conclusion]The results demonstrate that chain-of-thought prompting significantly outperforms few-shot prompting in extracting medication relationships.The experiments confirm the effectiveness of combining LLMs with knowledge graphs,showing that LLMs streamline the con-struction of knowledge graphs by enabling efficient information extraction,thus mitigating the complexities of traditional graph-building methods.Meanwhile,knowledge graphs offer structured,domain-specific knowledge for LLMs,enhancing the relevance and applicability of the generated answers.关键词
知识图谱/大语言模型/检索增强生成/智能问答/患者生成健康数据Key words
knowledge graph/large language model/RAG/intelligent Q&A/patient-generated health data分类
药学引用本文复制引用
彭佳儿,于菲菲,赵月华..从患者生成健康数据到用药建议生成[J].现代情报,2025,45(12):63-76,14.基金项目
国家自然科学基金资助项目"人智协同的多模态网络虚假健康信息识别及干预策略研究"(项目编号:72474098) （项目编号:72474098）
国家自然科学基金资助项目"基于深度学习的多源异构网络虚假健康信息识别研究"(项目编号:72004091) （项目编号:72004091）
中央高校基本科研业务费专项资金资助项目"基于大语言模型的多模态网络虚假健康信息特征感知及干预策略研究"(项目编号:2025300112). （项目编号:2025300112）
