| In the last few years,deep learning techniques have developed very rapidly in the area of natural language processing and have been employed extensively in areas such as text classification and intent recognition.Traditional deep learning-based text classification methods often require a large number of data samples with manual annotation to get better model classification results,and using manual annotation is time-consuming and laborious.In addition,in some fields,high-quality labeled data is not available in practical applications.In the absence of data,the models built by conventional deep learning methods have a series of problems such as poor generalization ability and overfitting,which cause the target accuracy of training requirements cannot be achieved.How to handle text classification tasks with few samples has become a popular research topic.By exploring and summarizing the current effective text classification methods and fewsample learning methods,this article approaches data augmentation based on generative adversarial networks from the perspective of few-sample data augmentation to generate diverse text sequences,solve the problem of too little training labeled data,and select a classification model applicable to few-sample datasets to improve the classification accuracy.Its main components include the following two points:(1)In terms of limited samples,choosing a suitable classification model is the key to improving classification efficiency and accuracy.In response to the fact that most current approaches focus on mining the shallow internal contextual information of text using static word vectors,this paper proposes an XLNet-based text classification model.The model combines XLNet pre-trained word vectors,a multi-headed attention mechanism,and an LSTM sequence model to capture the multi-faceted attributes of text information.At the same time,it can effectively apply the combination of deep learning model algorithm and GAN algorithm to few-sample text classification.(2)For the problem of few samples,with the idea of few-sample learning,the problem of insufficient samples under the condition of few samples can be effectively solved by using data augmentation techniques.In this paper,the GAN algorithm is used for text data augmentation,and a text classification model for few samples based on generative adversarial network is proposed.In the generative adversarial process,new samples can be generated by adding noise vectors to the vector of text feature representations and the text feature descriptions are optimised using the adversarial approach,resulting in semantically rich and diverse text with higher quality.Experimental results show that the proposed XLNet-based text classification model outperforms commonly used deep learning models in terms of classification performance.The text classification method with few samples based on GAN proposed in this paper also effectively improves the accuracy of text classification compared to other algorithms.These findings provide a valuable reference for solving the problem of few sample data in other fields. |
