Purpose/Significance To use artificial intelligence (AI) technology to quickly screen negative comments from a large number of reviews, so as to understand the needs and grievances of patients, and promote the sustainable development of telemedicine. Method/Process Taking comments from Haodf.com as an example, the paper first uses bidirectional encoder representations from transformers (BERT) to generate word embeddings, which are then fed into a convolutional neural network (CNN) and a bidirectional long short-term memory (BiLSTM) network in a dual-channel manner. Finally, a feature fusion strategy is employed to obtain textual sentiment information to achieve a binary classification task. Result/Conclusion The proposed dual-channel model based on BERT can better integrate the advantages of CNN and BiLSTM. It achieves the highest classification accuracy and macro F1-score compared to other 9 models, including BERT, BERT_BiLSTM, BERT_CNN, etc., which is effective in sentiment classification tasks for online medical reviews.