Automatic Adaptive Sequencing in a Webgame

摘要

Intelligent tutoring systems can improve student outcomes, but developing such systems typically requires significant expertise or prior data of students using the system. In this work we propose a new approach for automatically adaptively sequencing practice activities for an individual student. Our approach builds on progress for automatically constructing curriculum graphs and advancing a student through a graph using a multi-armed bandit algorithm. These approaches have relatively few hyperparameters and are designed to work well given limited or no prior data. We evaluate our method, which can be applied to a diverse range of domains, in our online game for basic Korean language learning and found promising initial results. Compared to an expert-designed fixed ordering, our adaptive algorithm had a statistically significant positive effect on a learning efficiency metric defined using in game performance.