Sentiment analysis is an important part of natural language processing. Students' learning status can be evaluated by analyzing their sentiment in online course comments. The CNN‐BiLSTM with attention model was established to evaluate the students' emotional state. This model can avoid the problem that convolutional neural networks ignore the contextual meaning of word, traditional recurrent neural network is easy to cause gradient disappearance and LSTMs can’t prioritize individual sections of the sentence that are relevant. The experiments on real online course comments showed that our proposed model outperforms three baselines in accuracy.