Student Engagement Detection in Online Classes

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering

Final Year Thesis Oral Defense

Title: "Student Engagement Detection in Online Classes"

by

WU Chi-hsuan

Abstract:

Online Learning is an interactive learning mode where students can acquire 
knowledge on various platforms without any distance barrier. With the 
widespread availability of internet services and the outbreak of COVID-19, 
online learning has drawn extensive attention. However, whether online 
classes are as effective as F2F classes remain questionable. Research has 
shown that during online courses, students often have shorter attention 
spans and lower concentration levels. Therefore, an engagement detection 
system will be essential to facilitate online learning effectiveness.

In this project, we utilized multi-level features to predict engagement 
scores and trained the model based on the idea of Momentum Contrast 
(MoCo). We explored that information from I3D models, Facial Action Units, 
and High-level behaviors (nodding, speaking, etc.) can have a 
complementary contribution to the prediction. With the combination of MoCo 
and margin loss, our designed training process can incorporate both 
ordinal relationships and the in-class variety of the label. Our work 
facilitates future engagement prediction related research.


Date            : 4 May 2023 (Thursday)

Time            : 09:30 - 10:10

Venue           : Room 5501 (near lifts 25/26), HKUST

Advisor         : Prof. CHENG Tim Kwang-Ting

2nd Reader      : Dr. XU Dan