Research on multi-label short text categorization method for online education under deep learning

The rapid development of the modern Internet has not only changed our way of life, but also changed the previous mode of education and learning, and the online education mode has been greatly developed and improved accordingly. In this paper, BERT model is used to extract word vectors of multilabel short texts for online education, and then BiLSTM-CNN model is used to extract features of short texts, and a classifier is constructed by Sigmoid activation function to realize the output of classification results of multilabel short texts. The validation analysis of the model’s effectiveness was conducted using the public dataset THCNEWS and the self-collected EduData as examples. The loss and Marco-P of the model after 5*10⁵ steps of training converged stably around 0.085 vs. 96.05%. The Marco-F1 values of the multi-label short text classification model on the THCNEWS and EduData datasets reach 0.915 and 0.962, which are significantly higher than the individual comparison models. Combining deep learning technology with multi-label short text classification for online education can achieve accurate classification of text data and provide new exploration ideas for improving the quality of online education.