Emotion recognition by deep learning to support mathematics learning and reduce math anxiety: A review of pedagogical approaches

Mathematics learning is affected by various achievement emotions such as enjoyment, boredom, frustration, and anxiety. Math anxiety can decrease motivation, perseverance, working-memory capacity, and performance while the teacher struggles to watch each individual student. This conceptual review consolidates research on mathematics-related emotions, emotion recognition based on deep learning, affective tutoring, and anxiety reduction with the help of artificial intelligence. This paper tries to answer two questions: (1) How can emotion recognition based on deep learning be used in math education to support learning and decrease anxiety? (2) What kind of technical, pedagogical, and ethical guidelines are necessary for using it? On the basis of the literature reviewed, a human-centric approach with three layers: sensing, interpretation, and pedagogical response is proposed. The visual, vocal, physiological, and interaction signals are processed into probability-based affect estimates which could be used for adjusting the pace, providing hints, scaffolding, and supportive feedback. Although it is possible to recognize emotions in a classroom, the performance of existing solutions depends on the task, dataset, emotion labels, and context and does not show any evidence of anxiety reduction. Thus, the role of emotion-aware artificial intelligence should be complementary to teacher’s judgments.