The Peer Data Labelling System (PDLS). A Participatory Approach to Classifying Engagement in the Classroom

Abstract

The paper introduces a novel and extensible approach to generating labelled data called the Peer Data Labelling System (PDLS), suitable for training supervised Machine Learning algorithms for use in CCI research and development. The novelty is in classifying one child’s engagement using peer observation by another child, thus reducing the two-stage process of detection and inference common in emotion recognition to a single phase. In doing so, this technique preserves context at the point of inference, reducing the time and cost of labelling data retrospectively and stays true to the CCI principle of keeping child-participation central to the design process. We evaluate the approach using the usability metrics of effectiveness, efficiency, and satisfaction. PDLS is judged to be both efficient and satisfactory. Further work is required to judge its effectiveness, but initial indications are encouraging and indicate that the children were consistent in their perceptions of engagement and disengagement.