Modeling Affect in Student-driven Learning Scenarios

Much research has been done on affect detection in learning environments because it has been reported to provide better interventions to support student learning. However, students’ actions inside these environments are limited by the system’s interface and the domain it was designed for. In this research, we investigated a learning environment wherein students had full control over their activities and they had to manage their own goals, tasks and affective states. We identified features that would describe students’ learning behavior in this kind of environment and used them for building affect models. Our results showed that although a general affect model with acceptable performance could be created, user-specific affect models seemed to perform better.