Research On The Generation Method Of Traditional Chinese Medicine Prescription Based On Generative Adversarial Networ

Traditional Chinese medicine(TCM)is one of the most important parts of Chinese culture.It has formed mature system of theoretical knowledge and retained plenty of data,which is valued to explore.To study the TCM data and inherit the traditional culture,we collected the data of herbs and prescriptions from Chinese pharmacopoeia and many classics about TCM.Then we did two research works on the basis of it.1)to know the clinical process and basic theories of TCM better,and learn how to generate prescriptions,we explored the data of TCM by data mining and other methods,including herbs and prescriptions.This work can also help us to know the relationships among attributes of data,such as prescriptions and herbs,symptoms and functions,etc.It was also helpful to the selection and preparation for the training data of neural networks.2)The generations and recommendations of prescriptions using GANs were also studied in this work.We learned characteristics and applications of different networks,then built appropriate models to do the experiment.The comparison experiments were also done.The details of these researches are as follows:1)The regulations among TCM data,including the composition of prescriptions,herb pairs,functions and symptoms,were explored by data mining and other methods.On the one hand,it is useful to understand and verify the classical theories of TCM,such as the process of clinical treatment and li-fa-fang-yao.On the other hand,it was helpful to the selections of data in training process of neural networks,such as the contents of training data and its format,etc.2)Through learning different networks,referring the structure of networks suitable for the generation of sequence,we explored the framework of model for the generation of prescriptions.The GAN was selected as the basic model to generate sequence of prescriptions.The format of training data,parameters and the networks of models were adjusted.Besides,to make generated prescriptions more specific,we referred CGAN to do conditional experiments.Different from previous works mostly applied to the generation of images,we need to combine the discreate sequence data with CGAN.In this work we need to adjust the structure of model and the format of training data with labels,such as functions or symptoms.Data augment also was used in this work to improve the scale and quality of training data.On the other hand,it can improve the generality of the model and avoid over fitting.Different form previous works,we were the first to proposed the views that prescriptions generations and recommendations using GAN and built the model to the experiment.These works not only can collect TCM data,verify its classical theories and explore its latent information,but also can gather information and inherent the traditional culture.Through the comparisons with generated prescriptions using LSTM,it further shows the feasibility and better effects on generating prescriptions using GANs.These works have advantages on finding the new prescriptions and the innovation of TCM.Moreover,it is helpful to provide assistance and reference to the clinical process online.