【目的】中药处方生成是智能中医研究中最具挑战性的课题之一。提出了一种用于中医处方生成的智能化模型——PreGenerator,它包含了一种新型的分级检索机制,可以自动提取处方和草药级模板,以实现临床准确的处方生成。【方法】PreGenerator首先使用症状-处方检索模块检索给定患者症状的最相关处方。为了遵循草药之间的配伍规律,引入草药-草药检索模块,根据前面生成的草药检索下一味最相关草药。最后,处方解码器融合症状特征和检索到的处方和草药的特征,生成预测中医处方。【结果】通过在真实医疗案例数据集上的自动评估和人工评估,验证了该模型的有效性。此外,模型可以推荐出一些没有出现在处方标签上但对缓解症状有用的草药。这表明该模型可以学习到草药和症状之间的一些相互作用。该研究也为未来传统中药智能查询和方剂生成的研究奠定了基础。

【Purposes】The generation of Traditional Chinese Medicine (TCM) prescription is one  of the most challenging tasks in the research of intelligent TCM. Although there is a small part of re‐ search in this field, transfer learning methods are usually used to apply relevant technology of text gen‐ eration to this task simply and roughly. Either large number of standardized dataset is needed to train  the model, or the domain knowledge and expertise of TCM are required. In order to solve these prob‐ lems, a hybrid neural network architecture for TCM prescription generation—PreGenerator is pro‐ posed. With a novel hierarchical retrieval mechanism, the PreGenerator can automatically extract pre‐ scription and herbal templates to facilitate accurate clinical prescription generation. 【Methods】First,  PreGenerator uses the Symptom-Prescription Retrieval module to retrieve the most relevant prescrip‐ tions for a given patient’s symptoms. In order to follow the rule of compatibility of herbs, the Herb- Herb Retrieval module is introduced to retrieve the next most relevant herb according to the condi‐ tioned generated herbs. Finally, the prescription decoder fuses the symptom features, the retrieved  prescription, and herbal template features to generate the most relevant and effective Chinese medi‐ cine prescription. 【Findings】 The validity of the model is verified by automatic evaluation and manual  evaluation on the real medical case dataset. In addition, the proposed model can recommend herbs that  do not appear on the prescription label but are useful for relieving symptoms, which shows that the  model can learn some interactions between herbs and symptoms. This research also lays a foundation  for the future research on intelligent query and prescription generation of TCM.