| With the development of the Internet and artificial intelligence,humanity has entered the digital society,resulting in a massive amount of text data.Effective management and utilization of this data can not only improve the information environment,but also bring great convenience to daily life.Therefore,text classification task has gradually become the focus of research in the field of natural language processing,and its applied algorithms have also changed over time from traditional machine learning methods to neural network models based on deep learning,as well as emerging attention mechanisms and pre training language models.In order to fully extract the Semantic information contained in the text and improve the classification accuracy,this paper uses the existing text representation technology and deep learning methods,and combines the BERT pre training model and neural network model to build a composite text classification model:BERT&LSTM&Text CNN(BLTN).The main tasks are as follows:(1)Construction of a dual channel neural network text classification model based on BERT.Due to the real-time,non-standard,and multiple implicit semantics of Chinese text,and considering that the Word2 Vec text representation method cannot generate text vectors rich in contextual information,a BERT pre training model is introduced to generate dynamic text word vector representations;On this basis,a dual channel neural network model is constructed based on the feature fusion idea to solve the problem that the traditional neural network model cannot fully extract semantic features:fuse the text local features extracted by the text convolution neural network(Text CNN)with the global text features extracted by the short and long term memory neural network(LSTM)containing contextual Semantic information to obtain more accurate and complete feature data.The THUCNews news headline text dataset was selected as the experimental corpus.Through comparative experiments with other deep learning models(Text CNN,BERT&Text CNN,BERT&Bi LSTM),it can be seen that the proposed model improves classification accuracy,accuracy,recall,and F1 value,verifying the effectiveness of the model.At the same time,the classification accuracy of each category in the dataset has reached a high level,indicating that the model has a certain degree of generalization.(2)To further validate the practicality of the BLTN model,it is introduced into the intention domain classification task in human-machine dialogue:understanding and analyzing user input based on contextual information,and categorizing it into a predefined set of intention domains to achieve intention recognition tasks and provide more accurate services.Through the analysis of the SMP Chinese human-machine dialogue technology evaluation dataset,it was found that the accuracy,accuracy,recall,and F1 value of the BLTN model classification reached over 95%.This result verified the feasibility and effectiveness of the BLTN model in human-machine dialogue,further demonstrating the practicality of the model. |
