A Deep Reinforcement Learning Based Emotional State Analysis Method for Online Learning

Abstract

: With the development of artificial intelligence technology, the basic judgment of students' learning state can be realized through the comprehensive analysis of students' face, expression, behavior posture and other multi-modal data. However, due to the lack of end-to-end recognition model and complete data sets, it is impossible to achieve accurate analysis of learning status. In this paper, based on deep reinforcement learning, an online learning state analysis method based on affective computing is proposed. On the basis of student identity recognition, face recognition is carried out through an unsupervised expression recognition model based on Siam-RCNN, and then 3D CNNs is used to recognize the feature data set for timing extraction. The state of collaborative awareness learning is analyzed by using HMM model. After verification, the accuracy of emotional state recognition can reach 98.88%, which is in the leading level in the industry.