Modeling the Affect of “Aha!" Moments to Detect the Moment of Learning

Abstract

What if a model could pinpoint the exact moment of learning? Currently, the only way we can understand when someone has learned is by testing them afterwards, which has its limitations. In attempts to detect the moment of learning, researchers from various fields have leveraged data from methods such as Knowledge Tracing (KT) and Electroencephalograms (EEGs) to predict students’ knowledge acquisition. These methods have contributed to improving our understanding of knowledge, but not only do they fall short of detecting the exact moment of learning, they also interfere with natural learning interactions by requiring students to wear sensors or type as they learn. Often, modeling learning does not include affect and emotion data, which are key influencers of learning outcomes. One affective expression that is often observed by educators, and has evaded quantification attempts by researchers, is the moment everything suddenly clicks for the student- the “Aha!” moment. Using classroom video data of students experiencing “Aha!” moments, we created dynamic, functional handcrafted features representing the face and body position and used them to model students’ facial expressions. We then leveraged feature selection methods and statistical analysis to ultimately contribute a novel, explainable definition of the observable, affective markers of “Aha!” moments, unlocking the opportunity to use the “Aha!” moment as a signal for detecting the moment of learning. These results invite future interdisciplinary research efforts as well as applications in fields such as artificial intelligence, human-robot interaction, education, psychology, cognitive sciences, and more.