AcuKG: a comprehensive knowledge graph for medical acupuncture

Abstract This study constructs an acupuncture knowledge graph (AcuKG) to systematically organize and represent acupuncture-related knowledge in a structured and scalable format. By extracting and integrating knowledge from diverse data sources, covering indication, treatment efficacy, practice guidelines, clinical research etc., AcuKG enhances knowledge discovery and utilization while improving data interoperability in the field of acupuncture. To achieve this, we employ multiple methods, including entity recognition, term normalization, semantic relation extraction, ontology mapping, etc., to extract and organize the acupuncture-related knowledge. Beyond knowledge structuring, to demonstrate AcuKG’s usability, we developed two use cases. First, AcuKG can assist obesity acupuncture knowledge discovery by linking key acupoints to obesity related acupuncture treatment efficacy research. Additionally, we demonstrate that knowledge injection into large language models (LLMs), such as ChatGPT and LLaMA, significantly improves their ability to answer acupuncture-related questions, increasing response accuracy. This work establishes a structured foundation for acupuncture knowledge representation, contributing to more reliable and efficient knowledge retrieval and discovery, and benefiting researchers, clinicians, and artificial intelligence (AI) applications in the field.