Manually Crafted Chinese Text Corpus for Text Emotion Recognition

Aiming at the scarcity of high-quality emotion corpus resources in the current Chinese text emotion recognition task, we have manually completed the annotation of 35,000 text data in seven typical emotion classes. At the same time, we also proposed a text emotion recognition model ERNIE-BiLSTM based on enhanced language representation with information entities (ERNIE) and bidirectional long short-term memory network (BiLSTM). The model can effectively extract text semantics and capture context information, which can greatly improve the capability of text emotion recognition. Finally, the model achieved good recognition performance on our text emotion corpus, with the accuracy, precision, recall, and F1 score reaching as high as 93.29%, 93.05%, 93.04%, and 92.98%, respectively.