Dissertation Talk: Applying Probabilistic Models for Knowledge Diagnosis and Educational Game Design
Seminar: Departmental | April 25 | 2-3 p.m. | Tolman Hall, Beach Room, 3105
Anna Rafferty, UC Berkeley
Educational technologies, such as massive open online courses and websites like Khan Academy, successfully deliver content to millions of learners. While these systems are able to effectively distribute educational resources, their ability to respond intelligently to learners is much more limited. In my dissertation, I consider how to use computational modeling and machine learning to interpret students interactions with educational technologies and personalize these technologies based on inferences about student knowledge. I will introduce a Bayesian inverse planning model for automatically diagnosing people's understanding based on their interactions with a game or virtual environment. Through behavioral experiments, I demonstrate that this model can accurately diagnose understanding from players' actions in a game, and show how the same algorithm can be adapted to make fine-grained inferences about people's algebra skills. I will then discuss how games or other interactive activities can be optimized in order to allow us to gain more information about a learner's understanding. This work highlights how computational models can be applied to educational questions in order to better understand human learning and create more effective educational technologies.