Semi-supervised model personalization for improved detection of learner's emotional engagement

• DOI: 10.1145/2993148.2993166
• Publication Date: 2016-11-01T09:46:03Z
• Authors: Nese Alyuz; Eda Okur; Ece Oktay; Utku Genc; Sinem Aslan; Sinem Emine Mete; Bert Arnrich; Asli Arslan Esme
• Keywords: Modalities

Affective states play a crucial role in learning. Existing Intelligent Tutoring Systems (ITSs) fail to track affective of learners accurately. Without an accurate detection such states, ITSs are limited providing truly personalized learning experience. In our longitudinal research, we have been working towards developing empathic autonomous 'tutor' closely monitoring students real-time using multiple sources data understand their corresponding emotional engagement. We focus on detecting related (i.e., 'Satisfied', 'Bored', and 'Confused'). collected 210 hours through authentic classroom pilots 17 sessions. information from two modalities: (1) appearance, which is the camera, (2) context-performance, that derived content platform. The platform consists section types: instructional where watch videos assessment solve exercise questions. Since there individual differences expressing engagement needs be customized for each individual. this paper, propose hierarchical semi-supervised model adaptation method achieve highly detectors. initial calibration phase, context-performance classifier obtained. online usage appearance automatically labels generated by model. experimental results show personalization enables performance improvement generic proposed result 89.23% 75.20% F1 measures sections respectively.