Research On Text Classification Based On BiGRU-CapsNet Model

With the continuous improvement of the level of science and technology,the Internet has developed rapidly.With the Internet as a link,various industries including commerce and trade,service industry,entertainment and leisure,public welfare and other industries have developed rapidly.People's lifestyles and working methods have undergone tremendous changes.Text,as the main bearing form of network information,its data volume is increasing rapidly and the fields involved are also more extensive.For example,the network has produced a large number of text information related to many industrial fields,such as movie reviews,news information,forum exchange information,microblog reviews,commodity reviews,etc.These texts have a huge amount of data and contain abundant information.After these texts are automatically classified,the real intentions of the information publishers can be understood,which is conducive to economic development,the direction navigation of leading enterprises in various industries,and the improvement of government decisions.In recent years,Deep Learning,as the most important progress in the field of artificial intelligence,has shown amazing performance in many fields.A large number of studies show that compared with traditional machine learning algorithms,many network models in deep learning can achieve better performance.The text classification model based on BiGRU network is currently the most mainstream text classification model,and has good performance in classification effect.This paper attempts to build a text classification model based on BiGRU model by using depth learning algorithm which is more suitable for text classification.The main work includes the following aspects:Firstly,the general process of text classification is summarized,including text preprocessing,text representation,text feature extraction,text classification training,text classification and performance evaluation.Through the research of the common methods in each step and the analysis of the characteristics of text classification itself,this paper further expounds many problems of the traditional methods of text classification and the difficulties of text classification,which lays a foundation for the selection of later classification methods and the design of classification network model.Secondly,the principle of BiGRU network and CapsNet network and their applications in text classification are expounded.In order to improve the text classification performance of the BiGRU network model,this paper deeply studies the neural capsule workflow and dynamic routing mechanism of CapsNet,analyzes the advantages and disadvantages of BiGRU network and CapsNet network in text classification respectively,and combines the bidirectional circulation mechanism of BiGRU with the neural capsule and dynamic routing mechanism of CapsNet to construct a text classification model based on BiGRU-CapsNet to classify the text.Finally,a BiGRU-CapsNet model is built on the keras framework platform to classify texts,and various factors affecting the performance of the BiGRU-CapsNet text classification model are discussed.Contrast experiments are respectively set up from three aspects:the number of iterations of the neural network,the length of intercepted text and the selection of activation functions,and the experimental results are analyzed to obtain the optimal parameters and function settings.On this basis,the model is used for classification training of other data sets to verify the applicability of the model,and the same data sets are tested and compared on BiGRU text classification model and BiGRU-CapsNet text classification model respectively.The experimental results show that,compared with the BiGRU text classification model,the BiGRU-CapsNet text classification model can classify text more effectively,resulting in higher classification accuracy and better applicability.