PRDAGE: a prescription recommendation framework for traditional Chinese medicine based on data augmentation and...

Area of Science:

• * Traditional Chinese Medicine (TCM)
• * Computational Medicine
• * Data Science

Background:

• * Traditional Chinese Medicine (TCM) prescriptions are vital for health maintenance and disease treatment.
• * Existing TCM prescription recommendation research primarily focuses on symptom-herb correlations, neglecting semantic information.
• * Limited dataset sizes in TCM research hinder effective model training and performance.

Purpose of the Study:

• * To address the limitations of current TCM prescription recommendation methods.
• * To develop a framework that captures and utilizes the semantic information of symptoms and herbs.
• * To improve the accuracy and generalization ability of TCM prescription recommendation models.

Main Methods:

• * Developed PRDAGE (Prescription Recommendation based on Data Augmentation and Graph Embedding) framework.
• * Created a dataset of 3,052 normalized classic medical cases.
• * Implemented a multi-layer embedding method using Sentence-BERT and graph convolutional networks.
• * Employed a median-based random data augmentation technique to enrich the dataset.

Main Results:

• * PRDAGE demonstrated superior performance compared to baseline models on an unenhanced dataset.
• * Achieved significant improvements in accuracy (1.69%) and recall rates (3.80%) at Top@10.
• * Ablation studies confirmed the positive contributions of both data augmentation and multi-layer embedding modules.

Conclusions:

• * PRDAGE is an effective framework for TCM prescription recommendation.
• * Multi-layer embedding successfully captures semantic information and complex relationships.
• * Median-based data augmentation enhances model performance and generalization.