Inferring Academic Emotion in Online Learning based on Spontaneous Facial Expression

Abstract

Academic emotion is one of the important factors impacting on learning effect. A robust automatic academic emotion inference method will have significant meaning for the educational field. This paper presents a study on the relationships between spontaneous facial expressions and academic emotions when students take online learning courses. First we establish corpora. The images from 82 students are collected with high definition cameras in an almost natural environment. Both students and external coders are invited to label the corpora. Then, the methodologies of academic inference algorithms are described based on both artificial feature and Convolution Natural Network (CNN). A preliminary experiment inferring self-annotation academic emotion is carried out to validate their actual effect. Among all the algorithms, the CNN-based algorithm using some tricks exhibited the ability to infer learners’academic emotion from the learner’s expression has the highest accuracy. This study has potential value to make up for emotion absence existing in online learning.