Visual Analysis of Medication-Symptom-Prescription Association in Traditional Chinese Medicine Ancient Texts

Ancient Chinese medical texts embody the foundational theories, pharmacological knowledge, and practical experiences of traditional Chinese medicine (TCM), possessing significant research value. Traditional methods of information extraction, filtering, and simple visualization of TCM ancient texts have failed to fully exploit the content and relevance of TCM theoretical knowledge. To address this issue, in close collaboration with domain experts, an interactive visual analysis method for exploring the correlation among herbs, symptoms, and prescriptions in ancient TCM texts is proposed. Leveraging a BERT model pre-trained on the corpus of the “Siku Quanshu”, a method for constructing and processing a TCM knowledge graph is designed. Based on the principle of “Jun-Chen-Zuo-Shi” in Chinese herbal medicine, a method for calculating the relevance between main symptoms and medicinal prescriptions is provided. An innovative layout of the knowledge graph for medicinal prescriptions is employed to support exploration of the correlation between TCM theoretical knowledge from the perspectives of herbs, symptoms, and prescriptions, rooted in the holistic view and differential treatment theory of TCM. Compared to traditional methods, this approach can better assist experts in efficiently exploring, understanding, and inferring from data in ancient TCM texts. Through comparative experiments, the proposed BERT_siku-BiLSTM-CRF model achieves precision, recall, and F₁ values of 90.57%, 93.53%, and 91.99%, respectively, for named entity recognition tasks, while the proposed BERT_siku-PCNN model achieves precision, recall, and F₁ values of 93.29%, 75.14%, and 80.40%, respectively, for relation extraction tasks, outperforming other benchmark models, thus validating the effectiveness of the proposed models. Two case studies on the application of “Compendium of Materia Medica” demonstrate the practicality of the interactive visual analysis system and receive positive feedback from domain experts during interviews.