<i>PreGenerator</i>: TCM prescription recommendation model based on retrieval and generation method

The generation of Traditional Chinese Medicine (TCM) prescription is one the most challenging tasks in research intelligent TCM. Current researches usually use transfer learning methods to apply relevant technology text this task simply and roughly. Either they need train a model with large number standardized dataset, or ignore domain knowledge expertise In order solve these problems, we propose hybrid neural network architecture for TCM generation— PreGenerator . It includes novel hierarchical retrieval mechanism, which can automatically extract herbal templates facilitate accurate clinical generation. Firstly, uses Symptom-Prescription Retrieval (SHR) module retrieve prescriptions given patient’s symptoms. follow rule compatibility herbs, Herb-Herb (HHR) introduced next herb according conditioned generated herbs. Finally, decoder (PreD) fuses symptom features, retrieved template features generate effective medicine prescription. validity verified by automatic evaluation manual on real medical case dataset. addition, our recommend herbs that do not appear label but are useful relieving symptoms, shows learn some interactions between This also lays foundation future query traditional medicine.