Affective Tutoring for Programming Education

抄録

This article discusses the use of artificial intelligence to detect student emotions while doing coding exercises for learning programming. Using data from programming students, we were able to build models for detecting confusion with as high as 70.46\% accuracy. We applied this in a system for programming practice that provides affective-based feedback by offering guides and adjusting the difficulty of exercises based on the presence of confusion, and found that students given affective feedback were able to solve more exercises and gave up less times. Finally, we also discuss the future direction of this research by collecting a larger amount of data that can cover other affective states and handle finer-grained detection of affect.