From May 28-30, 2018, Dr. Jaclyn Ocumpaugh had discussed the various processes used within the field and led a workshop emphasizing previously published research on educationally relevant affective states and the process of developing models.
Data mining has greatly expanded the opportunities for educational researchers to better understand student learning, especially in online learning systems. Researchers have explored constructs more basic to learning, with hundreds of articles exploring Bayesian Knowledge tracing (BKT), an algorithm that models the acquisition of specific skills in individual students. However, they have also modeled more complex processes, including long-term outcomes (passing annual standardized tests or even enrolling in college several years later).
As the number of students using these programs has increased, so has the feasibility of modeling more abstract components of the learning process, including educationally relevant emotional states (e.g. , boredom, confusion, engaged concentration, and frustration). Much of this research has relied upon sensors (video, audio, heart-rate, or posture sensors), but increasingly, researchers have sought to develop sensor-free models, which rely on student behaviors within the system to detect students’ emotions.