Research On News Text Classification Algorithm Based On Deep Learning

With the rapid development of Internet technology,people are gradually exchanging information on the Internet,so the amount of information on the Internet is growing rapidly.If the information is not classified,it will make it more difficult for people to get the information,such as making it more difficult for users to get the target news.Text classification can divide text information into different categories,so as to bring convenience for users to obtain information.The methods of text classification mainly include rules,statistics and deep learning,but the effect of rules and statistics depends on expert experience to a certain extent.In recent years,thanks to the efficient computing speed of computers,deep learning has developed vigorously and made great contributions to the development of text classification.This paper proposes five different deep learning-based models to solve the problem of news text classification.The main work of this paper is as follows:(1)A news text classification model based on cascading atrous convolution is proposed.The model mainly includes LSTM layer,cascade atrous convolution layer,pooling layer and full connection layer.Cascaded atrous convolution can improve the model performance by improving the model receptive field to extract the long-distance relations of text.(2)An improved news text classification model and its comparison model based on attention mechanism are proposed.The model mainly includes LSTM layer,grouped convolution layer,attention layer and full connection layer.The model integrates the different features extracted from the convolutional layer through the operation of feature combination after grouping the convolutional layer.The contrast model is mainly to change the attention layer to the pooling layer.By comparing the performance of the model with or without attention mechanism,it is found that attention mechanism is effective in improving the accuracy of news text classification.(3)A news text classification model based on atrous spatial pyramid pooling and its improved model are proposed.The two models mainly include LSTM layer,multi-scale atrous convolution layer,pooling layer and fully connected layer.The multi-scale atrous convolution layer of the general model consists of three groups of atrous convolution with different expansion rates,while the improved model includes a group of general convolution with a convolution kernel size of 1 in addition to the atrous convolution.Through experimental comparison,it is found that the improved model has better performance,indicating that adding the general convolution with the convolution kernel size of 1 can extract more information that is beneficial to classification.In addition,by comparing the performance of models with different convolution kernel sizes,it is found that whether the general model or the improved model,the model with convolution kernel size of 3 has better performance than the model with convolution kernel size of 2.In this paper,five different models are proposed and many experiments are carried out on open datasets.Compared with other advanced methods,the model proposed in this paper has a better effect,which indicates that the proposed method can effectively improve the accuracy of news text classification.Atrous convolution and atrous spatial pyramid pooling structure have been widely used in the field of image processing and have made great contributions to the development of image processing.In this paper,the atrous convolution model and the atrous spatial pyramid pooling model are introduced into the field of news text classification,and it is proved that they are effective in improving the accuracy of news text classification.This provides a new way of thinking and direction for the subsequent research on news text classification.