| With the development of internet technology,online media has become the main carrier of information,and news,as an important component of information,how to classify massive amounts of news has become a problem worth exploring.The application of machine learning in the field of text classification has gradually replaced traditional manual classification methods.Deep learning,as an important branch of machine learning,outperforms non deep traditional machine learning methods in processing massive amounts of information.Therefore,how to efficiently complete the classification task of online news through deep learning is of great significance for the development of automatic text classification technology.This paper uses a Mixture model combining Convolutional neural network(CNN)and Long Short Term Memory(LSTM)to classify online news texts,and improves the structure and Activation function of Convolutional neural network to improve the accuracy of the model in online news classification tasks.The main work done is as follows:(1)This paper summarizes the background and significance of the application of text classification technology in Natural language processing,analyzes the current research status of text classification at home and abroad and the related technologies of text classification,focuses on analyzing several text representation models,and puts forward a variety of optimization methods,which lays a good theoretical foundation for the follow-up training of God Network.(2)Aiming at the problems of slow training speed and low accuracy rate in traditional CNN text classification tasks,this paper proposes a method of combining convolution layer and BN layer to reduce the number of neural network layers,and replaces the Activation function with a smoother Mish function,so that the neural network can obtain more information to improve the efficiency and accuracy of classification.Through comparative experiments,it has been proven that the model using the Mish function performs better in text classification tasks.(3)A network model that combines CNN and LSTM is proposed to address the issue of information loss during feature extraction in the original neural network model.This model combines the advantages of CNN extracting local features of text and LSTM mining deeper semantic relationships of text.Word2 vec is used to convert network news text into word vectors,which are input into the combined neural network for training,Input the trained feature vectors into the fully connected layer for fusion,classify them using the Softmax function,and finally design a visual interface for the classification system.After experimental verification,the improved deep learning model has better classification performance in online news classification tasks,with an accuracy rate of94.04%. |
