【Purpose/significance】Sentiment analysis based on a large amount of UGC data has become an important task in public opin⁃

ion detection and social organizations. Emotion classification of WeChat tweets can provide a feasible management method for public

opinion dynamic regulation and public opinion trend monitoring. Most traditional classification models do not consider the contextual se⁃

mantic relationship of the text.【Method/process】This paper first combined CNN(the conventional neural network) and LSTM(long-term and short-term memory networks) to reconstruct the BiLSTM-CNN model, and then created a WeChat tweet sentiment classification model based on it.The model can get the emotional tendency of the text better, since the BiLSTM layer has given full consideration to contex⁃tual information.【Result/conclusion】Through the combination of different hyper parameters and model comparisons, the feasibility of the model is verified, and new theoretical models and methodological support are provided for the construction of WeChat tweets senti⁃ment classification models and method research.【Innovation/limitation】In the future work, we can classify the polarity of emotion in amore detailed way. In addition, because there are multiple bidirectional LSTMs in the BiLSTM layer, the training speed is very low and cannot be processed in parallel. We will study the methods to effectively accelerate the model training process.