Due to COVID-19, the teaching mode has shifted from in-person to blended or purely online. The level of engagement in class is one of the key factors that affect students’ overall learning outcomes. Therefore, it is essential for instructors to keep track of classengagement and adjust their teaching strategies accordingly (e.g., how fast they will go while giving online lectures). Traditionally, instructors could only evaluate students’engagement qualitatively (e.g. through occasional peeks). The discontinuity in evaluation is likely to not only result in unsatisfactory evaluation outcomes but also affect theinstructor’s teaching flow. To put this in a more concrete form, switching between teaching and evaluating would exhaust the instructor, and therefore negatively impact the overallteaching outcome. Now, with teaching shifting to online mode, it became even harder for instructors to identify students’ engagement levels due to the lack of face-to-face interaction. In the present state of the art, the computational models for students’ engagement exist but the use of contextual information (such as surroundings and text-based information) is very limited. Therefore, a context-aware deep learning basedstudent engagement detection is necessary for instructors to adapt to the current trend of online teaching. In this research, our team is interested in improving the context-based modeling of engagement for online educational scenarios. The research focus will be to determine the typical behaviors that suggest engagement or disengagement in online settings and configure the model accordingly to provide better accuracy. This specially-designed, highly-accurate engagement modeling is expected to assist instructorsin evaluating students’ engagement during online lectures, which will enable them to dynamically adjust their teaching method and personalize the course content based on the feedback from the model.
