Research And Application Of News Text Classification Based On Deep Learning

For many years,news text classification has been regarded as one of the key challenges in the field of natural language processing,which aims to classify text data into different categories.In practical applications,text classification is widely used in news aggregation,event detection,spam filtering,business decision-making,etc.At present,news text classification models based on CNN(Convolutional Neural Network)and RNN(Recurrent Neural Network)have achieved good classification results,but there are still some problems in these two text classification models,such as the inability to take into account the Lexical context information,insufficient text feature extraction,etc.Therefore,based on this model,this paper carries out the modeling of news text classification tasks,focusing on the combination of word vectors and the fusion of feature information in modeling,and constructs a news text classification system.The main contributions of the paper include the following aspects:(1)In view of the problem that the Glove model cannot take into account the vocabulary context information in the specific text,the ELMo and Glove dual-channel text classification model(ELMo-Glove-DCNN)was designed.ELMo generates dynamic word vectors,and Glove generates static word vectors.Splicing operation,using the DCNN model for local feature extraction,and then using the multi-head self-attention mechanism(Multi-head Self-attention)to dynamically adjust the feature weight coefficients for the features extracted by the DCNN model,and finally divide the news text into different parts through the Softmax function Classification.The experimental results show that the ELMo-Glove-DCNN model proposed in this paper has greatly improved the classification accuracy and improved the feature extraction ability of news text.(2)In view of the problem that ELMo and Glove only have language modeling and news text features are sparse during the feature extraction process,resulting in poor text classification effect,this paper proposes a DCNN,based on the ELMo-Glove-DCNN model.Bi GRU and multi-channel feature fusion text classification model(MC-FFTC)combined with multi-head self-attention mechanism(Multi-head Self-attention).The DCNN model combines specific target topic classification methods to extract local features of target keywords in text.Bi GRU is based on the analysis of sentence-level text features,extracts long sequence information,and finally performs feature fusion to construct a fused global feature vector.The experimental results show that the MC-FFTC model has greatly improved the evaluation indicators of Accuracy,Precision,Recall and F1 value because it solves the problem of gradient disappearance through the residual network,but it is slower than other models in terms of training time.long.(3)Based on the improved deep learning text classification model,this paper constructs an intelligent news text classification system.The requirements analysis,environment configuration,and detailed design and implementation of the system are carried out.