Leveraging part-and-sensitive attention network and transformer for learner engagement detection

Abstract

The assessment of student engagement levels in the context of online learning represents a pivotal challenge precipitated by the rapid and substantial progression of digitalization within the realm of education. This paper introduces a hybrid architecture, termed DTransformer, for engagement detection. Specifically, the architecture consists of two branches: a PANet and a STformer architecture. Extensive experiments are conducted on DAiSEE and EmotiW-EP dataset to obtain competitive performances for student engagement detection. The result shows that on DAiSEE dataset, the accuracy is 64% on the test set, whereas on the EmotiW-EP dataset, the MSE is 0.0729. The performances of our method has demonstrates the effectiveness of the proposed method.

Keywords

• Student engagement detection
• Deep learning
• Transformer
• 2D-CNN