To Kill a Student’s Disengagement: Personalized Engagement Detection in Facial Video

This work deals with engagement detection for e-learning systems based on a student’s facial video. We propose a novel, accurate technique that adapts the neural network-based binary classifier by using a short set of videos of a concrete user. The facial features are extracted with pre-trained lightweight deep networks that can be launched on a person’s device without a privacy leak. Moreover, we introduce the publicly available dataset that can be used to test the model’s efficiency. Experimentally, it is demonstrated that the approach with model adaptation can significantly (by 10–15%) improve the accuracy of predicting user engagement in the video.