Health query classification analysis based on BERT-TextCNN model

In order to automatically classify questions related to diabetes raised by patients, enhance the performance of search results, and promote the development of automatic diabetes question-answering services, this study adopts the BERT-TextCNN model (Bidirectional Encoder Representation from Transformers-Convolutional Neural Networks) for classifying diabetes-related health queries. The model is compared with BERT model, TextCNN model, BERT-TextRNN model, and BERT-TextRCNN model. The results show that, compared to individual models and other fusion models, the BERTTextCNN model achieves the highest precision, recall, and F1 score, with values of 0.8532, 0.8530, and 0.8506, respectively. In the validation phase, the BERT-TextCNN model also demonstrates higher accuracy, indicating superior classification performance.