Research on the Classification Method of CTC Alignment Joint-Test Problems Based on RoBERTa-wwm and Deep Learning Integration

Abstract

A text classification method based on RoBERTa-wwm and deep learning integration is proposed for fast classification of the massive unstructured problematic text data generated during the alignment joint-test of CTC (Centralized Traffic Control). Firstly, 10 common types of problems were summarized based on the statistical results of the problems from the alignment joint-test of CTC generated between 2011 and 2021; Secondly, in the text classification model, the pre-trained model RoBERTa-wwm is used to capture the semantic features of words in the problem text; Building an integrated model for deep learning based on BiLSTM-BiGRU and CNN to fully learn the deep hidden information in texts; Combining the principles of BiLSTM and BiGRU based on combined weight calculation to maximize performance; Normalization by Softmax function yields classification results for the CTC JCT problem. Finally, experimental validation is performed using data from CTC alignment joint-test problems generated during the last decade, The experimental results show that compared with several existing typical pre-trained models and classifier combinations, the proposed method in this paper achieves better results in terms of accuracy, precision, recall and F1 values, reaching 0.9317, 0.9322, 0.9317 and 0.9318, respectively, and the final Loss is only 0.24.