A Personalized Multi-region Perception Network for Learner Facial Expression Recognition in Online Learning

Learners’ emotions are closely related to their learning status. Effective emotion recognition helps educators grasp learners’ emotional states in a timely manner, thereby better adjusting teaching strategies and providing personalized guidance. Facial expressions, as the outward expression of inner activities, are an important basis for identifying human emotions. Therefore, facial expression recognition has become a common method for detecting learners’ emotional states, especially in online learning. However, the unique facial features of individual learners lead to different appearances of the same expression, which seriously affects the accuracy of facial expression recognition. To this end, we propose a personalized multi-region perception network (PMPN) for adaptively learning individual differences in facial expressions. On the one hand, a multi-scale global perception module (MGPM) is designed to capture feature representations at different scales to adapt to individual unique facial features. On the other hand, an adaptive local attention module (ALAM) is proposed to capture more personalized local features. Experimental results show that the proposed PMPN can accurately identify learners’ facial emotions in online learning by considering learners’ individual differences, and achieves the most significant recognition results. This method has positive significance for timely grasping the learning status of online learners, improving the online learning experience and realizing personalized education.