Key Finding
Integration of the AcuKG knowledge graph with large language models significantly improved accuracy in answering acupuncture-related questions, with GPT-4o improving from 46% to 54% and LLaMA 3 from 17% to 28%.
Researchers have created AcuKG, a comprehensive digital database that organizes all available information about acupuncture in one place. Currently, acupuncture knowledge is scattered across medical journals, clinical trials, websites, and traditional texts, making it difficult for both researchers and patients to find reliable, complete information. This new system brings together 1,839 different acupuncture-related terms and over 11,000 connections between them, linking information from PubMed studies, clinical trials, and established medical databases. The researchers tested their system in two ways: first, they used it to identify which acupuncture points work best for obesity treatment, successfully pinpointing key points like ST25 and ST36 based on scientific evidence. Second, they combined their database with artificial intelligence chatbots (like ChatGPT) to answer acupuncture questions more accurately. When the AI had access to this organized acupuncture knowledge, its accuracy improved significantly—GPT-4o went from 46% to 54% correct answers, while another AI called LLaMA 3 improved from 17% to 28% accuracy. For patients, this means that healthcare providers and researchers now have better tools to understand which acupuncture treatments have scientific support for specific conditions. It also means AI-powered health information tools can provide more reliable acupuncture guidance when they use this structured knowledge base. As acupuncture gains recognition worldwide as a complementary therapy, having organized, evidence-based information becomes increasingly important for making informed treatment decisions. If you're considering acupuncture, consult with a licensed acupuncturist who can assess your individual needs and recommend an appropriate treatment plan.
This study describes the development of AcuKG, a knowledge graph integrating acupuncture data from multiple sources including PubMed, ClinicalTrials.gov, and medical ontologies (SNOMED CT, UBERON, MeSH). The database contains 1,839 entities and 11,527 relations mapped to 1,836 standardized concepts. Methodology employed entity recognition, relation extraction, and ontology mapping with human-in-the-loop quality control. Two validation cases assessed clinical utility: (1) obesity research application identifying evidence-based acupoints (ST25, ST36), and (2) integration with large language models for clinical question-answering. LLM performance improved significantly when augmented with AcuKG—GPT-4o accuracy increased from 46% to 54% (P=0.03), and LLaMA 3 from 17% to 28% (P=0.01). Clinical takeaway: AcuKG provides practitioners with a structured, evidence-based framework for point selection and represents a significant advancement in integrating traditional acupuncture knowledge with modern computational tools, potentially improving clinical decision-making and patient education through AI-enhanced applications.
Browse our directory of verified licensed practitioners near you.
Find a practitioner →📌 Combined Banxia-Baizhu-Tianma Decoction and acupuncture shows preliminary evidence of superior outcomes compared to monotherapy for cerebral ischemic stroke, improving cerebral perfusion, motor function, and reducing disability rates through synergistic targeting of overlapping pathophysiological pathways.
📌 Traditional Chinese Medicine interventions, including herbal formulations and acupuncture, effectively enhance thyroid function, decrease autoantibody levels, and improve quality of life in Hashimoto's thyroiditis patients through immunomodulation and reduction of oxidative stress.
📌 Metabolomics reveals that acupuncture and herbal formulas consistently modulate amino acid, lipid, and energy metabolism pathways across multiple neurological disorders, providing quantitative evidence for TCM's multi-target therapeutic mechanisms.