Learning Individual Behavior in an Educational Game: A Data-Driven Approach.
Seong Jae Lee, Yun-En Liu, Zoran Popović · 2014
In recent years, open-ended interactive educational tools such as games have been gained popularity due to their abil-ity to make learning more enjoyable and engaging. Model-ing and predicting individual behavior in such interactive environments is crucial to better understand the learning process and improve the tools in the future. A model-based approach is a standard way to learn student behavior in highly-structured systems such as intelligent tutors. How-ever, defining such a model relies on expert domain knowl-edge. The same approach is often extremely difficult in edu-cational games because open-ended nature of these systems creates an enormous space of actions. To ease this burden, we propose a data-driven approach to learn individual be-havior given a user’s interaction history. This model does not heavily rely on expert domain knowledge. We use our framework to predict player movements in two educational puzzle games, demonstrating that our behavior model per-forms significantly better than a baseline on both games. This indicates that our framework can generalize without requiring extensive expert knowledge specific to each do-main. Finally, we show that the learned model can give new insights into understanding player behavior.