Academic emotions of learners are important for academic achievement. For the online learning platform, it is of great value to gain insight into the academic emotion of the course in appropriate time interval from the platform. We crawled a large number of student comment texts from MOOC, and used deep learning algorithms (BERT models) to perform aspect-oriented sentiment classification on the comment texts. We conducted statistical analysis and identified keywords to explore the changes of academic emotions in the online learning environment in different aspect dimensions. The results show that academic emotions are significantly improved in the first and second period of the course schedule, and tend to be stable in the second and third period of the course schedule. From the word frequency statistics, in the dimension of the teacher, students' concerns mainly focus on two aspects: One is whether they can acquire knowledge, the other is the characteristics of teachers; in the course dimension, students attach more importance to the learning; in the dimension of the platform, students' negative emotions mainly focus on four aspects: certificate, learning record, prompt and subtitle. Our research aims at providing suggestions for course design, platform improvement, and teachers' practice.