Automatic Academic Emotion Recognition and Its Evolution Analysis in Classroom

Students' emotional state plays a vital role in learning. In this study, we examined the emotional changes of 69 students during the lectures of the C Language Programming Foundation and Computational Thinking in central China. This paper presents a method using Yolov7 and the convolutional neural network CoAtNet to monitor students' emotional evolution by analyzing their facial expressions and recognizing academic affective states, including boring, confused, focus, happy, and neutral. The trained model has achieved a test accuracy of 76.0%. The results showed that focus increased at the beginning of the first 10 minutes of class and then began to decline. With the decline of focus emotion, students' neutral emotions increased. We found that when a new teaching session is performed, or the teaching method of naming questions is used, the students' emotion of focus increases briefly, lasting about 8–10 minutes. Subsequently, even if the varied teaching method was used to stimulate students' attention, their level of engagement did not increase significantly. It can be suggested that teachers can use teaching methods appropriately to promote students' attention. However, students' attention span is limited, and teachers should pay attention to methods and time to improve teaching quality effectively.