融合双通道特征信息的医疗短文本分类模型

打开文本图片集

Medical short text classification model with fusion of dual channel feature information

LI Chen²，LIU Na1，2 ，ZHENGGuofeng1，²，YANGJie1，²，DAOLu1，² （1.CollegeofComputerScienceandEngineering，NorthMinzuUniversity，Yinchuan75o021，China; 2.TheKey

Abstract：Inviewofthesparsefeatures，semanticambiguitiesandinsuficientextractionofshorttextfeaturesinthe medicalshorttexts，amedicalshorttextclassificationmodelEBDF（ERNIE-BiLSTM-DPECNN-FGM）fusingdual-chanel featuresisproposed.Thepre-trainedmodelisusedtoobtaindynamicwordvectors，whichmadethemodelcontainricher semantic information.Thenthe BiLSTMisusedtoobtainglobaltextfeature informationandthe DPECNNisusedtoobtain deep localtextfeatureinformation.TheFGMadversarialtrainingalgorithmisusedtodisturbancethedatatoimprovetherobustne andgeneralizationabilityofthemodel.Finally，thefeatureinformationofthetwochannelsisfusedtoobtainthefinaltext representation.TheEBDFmodelwascompared with the model withthe better efectontheshorttextdatasetsof three medical fieldsand two general fields.It can be seen that itsaccuracyis improved by about 0.57%＼～6.16%，and its F1 value is improved by about 0.65% 3 5.80%

Keywords：medical text mining;short text clasification;feature fusion;BiLSTM; DPECNN;two-channel

0 引言

医疗行业是一个数据密集型和知识密集型的行业。（剩余16371字）