Abstract:Automated measurement of student engagement equips educators with valuable insights, aiding them in achieving educational program objectives and customizing their approach to suit individual students. Engagement measurement requires a detailed analysis of the behavioral and affective states of students over precise timescales. A range of current techniques have engineered sequential and spatiotemporal models, including recurrent neural networks, temporal convolutional networks, three-dimensional convolutional neural networks, and transformers to measure engagement from video data. These models are trained to incorporate the sequential/temporal order of behavioral and affective states into the video analysis, outputting their level of engagement. Drawing upon the definition of engagement in educational psychology, this paper questions the necessity of incorporating the order of behavioral and affective states into engagement measurement. Non-sequential bag-of-words-based models are developed to analyze behavioral and affective features extracted from videos and output engagement levels. The non-sequential models only analyze the occurrence of behavioral and affective states not the order in which they occur. Experimental results indicate that the proposed non-sequential approach is superior to state-of-the-art sequential engagement measurement approaches. On the IIITB Online SE dataset, the proposed approach significantly improved engagement level classification accuracy by 22%, and 26%, respectively, compared to the recurrent neural network, and the temporal convolutional network. It also improved minority class recall and achieved a classification accuracy as high as 0.6658 On the DAiSEE dataset. In another experiment, models displayed consistent performance while trained on the shuffled versions of the datasets compared with those trained on the original, unshuffled datasets. In the shuffled versions, behavioral and affective states within video samples were randomly permuted. These observations reinforce the notion that the order in which affective and behavioral states occur does not impact engagement measurement.

Bag of states: a non-sequential approach to video-based engagement measurement

Personalized User Engagement Modeling for Mobile Videos

Engagement Detection in Online Learning Based on Pre-trained Vision Transformer and Temporal Convolutional Network

Engagement Measurement Based on Facial Landmarks and Spatial-Temporal Graph Convolutional Networks

Detecting disengagement in virtual learning as an anomaly using temporal convolutional network autoencoder

DMCNet: Diversified Model Combination Network for Understanding Engagement from Video Screengrabs

Improving state-of-the-art in Detecting Student Engagement with Resnet and TCN Hybrid Network

A General Model for Detecting Learner Engagement: Implementation and Evaluation

Multimodal Engagement Analysis from Facial Videos in the Classroom

Measuring Student Behavioral Engagement using Histogram of Actions

Enhancing frame-level student engagement classification through knowledge transfer techniques

An intelligent system for monitoring students' engagement in large classroom teaching through facial expression recognition

Evaluating Children’s Engagement Through Wrist-worn Biosensors and LSTM Network

Class-attention video transformer for engagement prediction

Predicting Engagement in Video Lectures

Are you still with me? Continuous Engagement Assessment from a Robot's Point of View

Multimodal Engagement Recognition From Image Traits Using Deep Learning Techniques

Class-attention Video Transformer for Engagement Intensity Prediction

Engagement Estimation in Advertisement Videos with EEG

Delving Deep into Engagement Prediction of Short Videos

Prediction and Localization of Student Engagement in the Wild