Research And Application Of News Text Classification Method Based On Deep Learning

With the rapid development of Internet and big data industry,news has become one of the important means for people to understand social dynamics and obtain social information resources.News text classification is helpful to the management of news information,the realization of news order and the mining of news data.It is of great significance to save human resources and obtain valuable news information efficiently.However,the current research on news text classification is mostly in English,and there is a lack of news text classification and related corpus in Chinese.Moreover,the traditional machine learning methods commonly used in long text processing have problems such as imperfect text feature extraction.The emergence of deep learning has further broken through the bottleneck faced by machine learning and brought significant opportunities to the field of text classification.For this reason,this paper introduces deep learning algorithms into the field of news text classification research and carries out the research and application of news text classification methods based on deep learning.The specific work is as follows.Firstly,in the aspect of text representation,there are the problems of less Chinese news text classification and lack of related corpora.According to the constructed data index,this paper uses weight extraction and word frequency statistics to create a vocabulary suitable for Chinese long text classification,and use One-hot and Word2 Vec tools to embed the word vector of text data,which provides a convenient word vector mapping for the input and extraction of text features.Secondly,this paper proposed a technical framework for Chinese news text classification,and improved on the basis of the deep learning model Gated Recurrent Unit and Convolutional Neural Network,respectively proposed and implemented a hierarchical and bi-directional GRU model and the combined CNN model.The hierarchical and bidirectional GRU model realizes the effective memory of the information before and after the long text by the way of forward and reverse bi-directional channel and multi-layer hidden layer information transmission.The combined-CNN model makes the local feature extraction of text blocks more comprehensive by convolution pooling and combination.The experimental results show that the hierarchical and bi-directional GRU and the combined CNN classification algorithms improve the accuracy of Chinese news text classification,and the accuracy rate is 93.20% and 93.69% respectively,which is better than the traditional machine learning algorithm and deep learning algorithm.After the data set is balanced,the improved model further achieves better classification results,the accuracy rate is 95.80% and 95.57%respectively,and the recall rate and F value also achieve good results.Thirdly,to prevent the problems of overfitting,gradient disappearance and gradient explosion encountered in the training of machine learning models and deep learning models,effective model regularization and RAdam optimization algorithms are designed in this paper.The regularization method obtains the appropriate number of model iteration rounds and introduces Dropout layer to alleviate the overfitting phenomenon.The optimization algorithm is compared by experimental effects,and the emerging RAdam optimization algorithm is used to optimize the model training in an adaptive learning rate way.Finally,on the basis of realizing the automatic classification of news texts,the hierarchical and bidirectional GRU and combined CNN classification algorithms are applied to the news classification personalized recommendation system,and a visual news interface is designed and developed.The system uses We Chat small program as the implementation platform,and establishes the application of news reading and music recommendation.The application system solves the information overload problem caused by massive news data,and satisfies people's habit of listening to music when reading news.In summary,the theoretical study and experimental analysis show that the research work of this paper has significant effect on news text classification improvement,which has extremely significant economic and social benefits.