Affect Detection in Home-Based Educational Software for Young Children

Research on automated affect detection in educational software using play log data has shown promising results. Yet most studies use classroom-based software designed for adolescents or adults. In this paper, we aim to detect affection in an online educational platform primarily aimed at home use by young children. This presents two challenges: we have to rely on a self-report instrument of affect that users can utilize at home, and we have to make sure that this instrument is properly understood by children. To this end, we developed and validated an emoticon-based self-report instrument to derive ground-truth labels of four emotions: Joy, frustration, confusion, and boredom. Training a number of different classifiers for automated affect detection yields promising results, in particular for detecting joy and frustration. [For the full proceedings, see ED599096.]