Research on CGAN-BERT and RGAN-BERT Models for Short Text Classification based on Semi-Supervised Model

With the rapid development of artificial intelligence, a large number of short texts cause a certain degree of information redundancy. Text classification technology can help people classify and process information, and has important applications in the fields of recommendation system, public opinion monitoring and information retrieval. However, short text information in different fields has the characteristics of industry professionalism and fast updating of language style. The resulting problems such as poor applicability of the model and bottlenecks in annotation make the effect of traditional classification methods in short text analysis limited. Therefore, we propose semi supervised text classification models CGAN-BERT and RGAN-BERT based on convolutional neural network and cyclic neural network respectively. The newly designed generator and discriminator are more conducive to the game process. We evaluate our model and other classical models on several public data sets. The experimental results show that our proposed model is better than other models.