Study On Mining Herb Pairs Of Traditional Chinese Medicine Based On ERNIE Model

Herb pairs is a relatively fixed combination of decoction pieces commonly used in clinic.Its compatibility has certain regularity.Mining herb pairs is to find a certain degree of coincidence or restricted combination of slices from many decoction pieces.In view of the low accuracy and the lack of attribute information of mining herb pairs,this paper combined with natural language analysis technology and focuses on the compatibility of herb pairs.The specific contributions are as follows:1.In order to improve the training effect of the herb pairs mining,the method of expanding herb pairs data was put forward to solve the problem that the data of herb pairs was published less.In that present invention,the text generated quality of herb pairs by the generative adversarial networks is low,so a model of Generation Adversarial Network Based on Hierarchical Learning with Multi-reward Text is proposed.It adopts a hierarchical structure,and using the transfer feature vector training module which is transferred to the generator by leakage of information.Therefore,the accuracy of long text generation is improved.In addition,Relational Memory Core is used as the bottom training module of the hierarchically generate adversarial network.It allows the self-concern mechanism to interact in the memory,so that enhancing the expression ability of the generator and the discriminator.Lastly,the sentence-word reward mechanism is adopt to send the feedback information of the discriminant back to the generator,which is used to guide the generation of the herb pairs data,The results show that the proposed model can not only make the generated herb pairs to the text more accurately,but also reflect the time complexity of the model better than the existing reward mechanism.2.The shallow machine learning algorithm for mining herb pairs focuses on the frequency of occurrence.This paper comprehensively considers the co-occurrence information among the decoction pieces,and counts the frequency of all kinds of information on the nature,taste,channel tropisms and functions of the potential herb pairs.ERNIE＿DPCNN model is used to mine herb pairs,because the model can extract the comprehensive potential characteristics of different decoction pieces in the aspects of nature,taste,channel tropisms and functions.So that the deep correlation between herb pairs can be effectively discovered.By comparing the real data set with the existing mining herb pairs model,the results show that the model can further improve the accuracy of searching herb pairs and mining potential herb pairs.The approaches and corresponding technologies proposed in this thesis can be applied into several real application fields such as TCM prescription,herb pairs analysis,prescription compatibility,text expansion,data enhancement,text diversity generation and other application fields,which play an important role in improving the quality of existing systems in the above fields.The dissertation includes 21 figures,9 tables and 72 references.