Research On Text Classification Algorithm Based On Deep Learning Method

Text classification has always been one of the hot topics in the field of natural language processing.In the application of search engine,intelligent customer service,intelligent robot and so on,how to understand the meaning of text sentence according to the context is a key problem,but also a difficult problem.Traditional text classification mainly adopts regularization method based on rules and template matching or traditional machine learning method,but it has the problems of high computational cost and poor generalization ability.In order to solve the above problems,this thesis proposes a transfer learning method based on BERT-FNN model to solve the problem of text classification and conducts a comparative experiment.The main research contents and achievements are as follows:(1)Based on the traditional machine learning method to solve the problem of text classification.In this thesis,two machine learning methods,LR and SVM,are used to complete the task of text classification.The research shows that the accuracy rate of LR method is 91%,compared with SVM method,the accuracy rate,recall rate,F1 value and other performance are significantly improved.(2)Based on the traditional deep learning method to solve the problem of text classification.In this thesis,CNN and LSTM are used to complete the task of text classification.The research shows that the accuracy of LSTM method is 92%,compared with the traditional method,the accuracy,recall,F1 value and other performance are significantly improved.(3)Transfer learning method based on BERT-FNN model for text classification.based on Google’s public best pre training language model,the input text context modeling and sentence level semantic representation are carried out,and then the sentence features are extracted by fully connected neural network(FNN).Therefore,a text classification method of transfer learning based on BERT-FNN model is proposed.This algorithm transfers the text features obtained by pre-training to the target domain text classification,then trains the model on the target domain,and fine-tuning the parameters.Through this mechanism,the accuracy of text classification is improved when the target task data is insufficient.Compared with LR,SVM,CNN,LSTM,BERT and BERT-FNN models,the experimental results show that the accuracy of this method to solve text classification problem reaches 95%.Compared with other traditional methods,the accuracy,recall rate,F1 value and other performances are significantly improved.This shows that the performance of the transfer learning method based on BERT-FNN model is better than the traditional model in text classification task,and the effect is more significant.