Research On Text Classification Model Based On Dynamic Representation

Text classification is to automatically classify text sets according to certain classification methods or standards.Text classification is a basic task in the field of natural language processing,and also a very important task.As one of the most classic NLP scenarios,the study of text classification has great practical significance.Improving the accuracy of text classification in the application field can greatly save more manpower investmentAiming at the problem of static representation of word vectors,this paper explores a dynamic representation model based on BERT structure.First of all,we collect,analyze and build vocab for the data set,then extract the corresponding word vector from the word vector based on the training of Sogou news data set according to the corresponding vocab,then study and improve the excellent text classification models TextCNN,FastText,TextRNN,TextRCNN in recent years,and build Dynamic TextCNN,Dynamic FastText,Dynamic TextRNN and Dynamic TextRCNN model.The results show that:(1)FastText is better than TextCNN,TextRNN,TextRCNN and other models in the news title text classification data set;(2)the accuracy of Dynamic TextCNN in the test set is increased from 91.53%to 95.37%;(3)the accuracy of Dynamic FastText in the test set is increased from 92.65%to 95.26%;(4)the accuracy of Dynamic TextRNN in the test set is increased from 91.05%to 95.21%;(5)the accuracy of Dynamic TextRCNN in the test set is increased from 91.59%to 95.34%.Finally,the deep learning text classification model constructed in this paper is supported by the current domestic text classification model methods,which has certain reference value for NLP related tasks based on dynamic representation.