Privacy-Preserving and Scalable Affect Detection in Online Synchronous Learning

Abstract

The recent pandemic has forced most educational institutions to shift to distance learning. Teachers can perceive various non-verbal cues in face-to-face classrooms and thus notice when students are distracted, confused, or tired. However, the students’ non-verbal cues are not observable in online classrooms. The lack of these cues poses a challenge for the teachers and hinders them in giving adequate, timely feedback in online educational settings. This can lead to learners not receiving proper guidance and may cause them to be demotivated. This paper proposes a pragmatic approach to detecting student affect in online synchronized learning classrooms. Our approach consists of a method and a privacy-preserving prototype that only collects data that is absolutely necessary to compute action units and is highly scalable by design to run on multiple devices without specialized hardware. We evaluated our prototype using a benchmark for the system performance. Our results confirm the feasibility and the applicability of the proposed approach.