Research And Application Of Text Classification Based On Deep Learning

With the rapid development of mobile Internet,complex information appears in people’s life in various forms.As one of the main carriers of information,text plays the role of information transmission bridge in life.Nowadays,in the face of massive data,relying on computers to process text has become an excellent and necessary choice.As one of the most basic tasks in the field of Natural Language Processing(NLP),Text Classification plays a huge role and has far-reaching influence.From the initial text classification algorithm based on emotion dictionary to the algorithm based on machine learning and deep learning,text classification task has experienced rapid and profound development.Text classification task has a wide range of applications and huge value,so it is particularly necessary to build a scientific and reliable text classification model.This paper mainly uses the method of deep learning to fuse and improve the current text classification model,focusing on improving the accuracy and F1 value of text classification,so as to obtain a model with good performance,and verify its classification effect by applying it to multiple data sets through experiments.The main work and innovations of this paper are as follows:1.Firstly,the relevant data set used in this paper is introduced.At the same time,the process of text preprocessing is illustrated by taking the data set of network film review as an example,and the processing results of relevant steps are given.Then,four main evaluation indexes of text classification task are introduced: accuracy rate,accuracy rate,recall rate and F1 value.Then,in view of the problem that there is a certain improvement space for the previous evaluation indicators such as text classification,accuracy of sentiment analysis algorithm and F1 value,this paper integrates convolutional neural network(CNN)and bi-directional gated cyclic unit(BiGRU),and combines the local feature extraction of convolutional neural network convolution layer.And the bidirectional gated loop unit simultaneously focuses on the advantages of feature extraction from the two input directions before and after the text,and summarizes the problems that others use BiGRU layers less,leading to incomplete feature extraction,and too many BiGRU layers,leading to slow training speed.Through experiments,appropriate multi-layer BiGRU is obtained.It can not only improve the accuracy of text classification,F1 value and other aspects,but also ensure proper training time of the whole model through appropriate multi-layer BiGRU.After connecting the whole connection layer,it can output emotion polarity through sigmoid function,and design CMBG model based on Word2 vec word vector.Experiments on online film review data set and hotel review data set show that the proposed model has made some progress compared with the previous model.2.Sentiment analysis is a binary classification of positive and negative text emotions,but there are multiple classification tasks such as news categories.In this chapter,a TCNNn(Transformer+CNN+n Adam)model based on improved Adam optimizer and integration of Transformer mechanism is proposed.By learning the attention mechanism and calculating the contribution of input data,the autocorrelation of each word in the text is established.Combined with CNN’s feature of extracting local features,a classification experiment is carried out on the news text data set.In addition,in order to further verify the effectiveness of the model,ablation experiment was also carried out at the end of this paper.The results of both the comparative experiment and ablation experiment confirmed that the model did improve the text classification evaluation index compared with the baseline model,which verified the effectiveness of the model.