A survey on data-driven approaches in educational games

Abstract

Open-ended educational systems such as games have been broadly under investigation in recent years due to their potential in making learning enjoyable and offering the adaptive pedagogy of intelligent tutoring systems. The most important challenge in building such systems is to predict individual behavior which results in better understanding of the learning process. Model-based methods are a standard way to learn individual behavior in highly-structured systems. However, these methods heavily rely on expert domain knowledge. Since adaptive educational games may create a huge space of actions, applying model-based approaches in these systems are very difficult. In order to counter this difficulty, researchers utilize data-driven methods that are not dependent on expert domain knowledge to learn a subject's behavior based on a history of user interactions. Due to the fact that the potential of applying data-driven approaches in adaptive educational games is still missing, the goal of this report is to cater a survey in the area in order to ease comprehending the state of the art.