Abstract: Emotion recognition in online Chinese educational reviews is largely limited by a lack of annotated data. To solve this problem, this paper constructs a Chinese MOOC emotion recognition corpus from 806 college MOOCs in China by combining automatic and manual methods to ensure corpus balance and extensive subject coverage, including a total of 10 340 reviews, of which 5 411 are positive and 4 929 are negative. Firstly, it formulates strategies of corpus collection and pre-processing, annotation specification, annotation system and consistency detection method. Then, a neural network model and an emotion recognition method based on large-scale pre-trained language models are proposed. Finally, the results of emotion recognition are applied to teaching management department and instructors. The corpus lays a data foundation for emotion analysis research of online education reviews, and is of great significance for enabling teaching evaluation and facilitating the intelligent teaching system.