LLM-KG协同优化：睡眠障碍管理新范式

打开文本图片集

中图分类号：TP183.4 文献：

文献标识码：A

LLM-KG Synergistic Optimization： A New Paradigm in Sleep Disorder Management

ZHAO Chenjun，ZHAO Aite （College of Computer Science and Technology，Qingdao University，Qingdao 266O71，China）

Abstract： Aiming at the lack of specialized knowledge graphs in the field of sleep medicine and the " halucination" problem in large language model question answering， this paper proposes the Sleep-centric LLM-KG Collborative Question Answering System （SLKG-QA），which achieves the integration of dynamic knowledge expansion and reliable question answering through the complementary mechanism between the Knowledge Graph （KG） and the Large Language Model （LLM）. This model consists of a question-answering phase and a knowledge updating phase. In the question-answering phase， the system utilizes the knowledge graph to perform real-time retrieval and constraints on the LLM's generation process，effectively suppressing factual erors. In the knowledge updating phase，leveraging the semantic understanding capability of the LLM， the system automatically identifies and supplements missing or outdated triples in the knowledge graph，enhancing its coverage and timeliness. Experimental results show that the accuracy of the sleep medicine knowledge graph constructed based on this method exceeds 90% ，with domain specificity reaching approximately 85% . Compared with existing methods，both question-answering accuracy and domain specificity are improved by over 20% ，validating the effectiveness of this collaborative model in achieving dynamic knowledge expansion and ensuring the reliability of question answering.

Keywords： deep learning;knowledge graph; large model; sleep disorder

随着现代生活节奏的日益加快，睡眠相关疾病已成为全球性健康挑战，其高患病率、强隐蔽性及多并发症等特点严重威胁个人健康[1]。（剩余12028字）