Discovering Knowledge Models in an Open-ended Educational Game using Concept Formation

Developing models of the knowledge and skills being exercised in a task is an important component of the design of any instructional environment. Developing these models is a labor intensive process. When working in exploratory and open-ended environments (EOLEs) the difficulty of building a knowledge model is amplified by the amount of freedom afforded to learners within the environment. In this paper we demonstrate a way of accelerating the model development process by applying a concept formation algorithm called TRESTLE. This approach takes structural representations of problem states and integrates them into a hierarchical categorization, which can be used to assign concept labels to states at different grain sizes. We show that when applied to an open-ended educational game, knowledge models developed from concept labels using this process show a better fit to student data than basic handauthored models. This work demonstrates that it is possible to use machine learning to automa...

• Type: Journal
• Year: 2015
• Where: AIED
• Authors: Erik Harpstead, Christopher J. MacLellan, Vincent Aleven
• Keywords: AIED 2015, Artificial Intelligence