Serious games that should adapt training to the individual might benefit from methods that are developed for intelligent tutoring systems. One method, model tracing, might be used for domains with a strict hierarchy, while complex domains that lack such a hierarchy might use the method we introduce in this paper as teacher modeling. It takes the perspective of a teacher, who does not always know what leads a student to an answer, but who does know when the answer is (in-) correct and who can assess the capacities of the student over time. The assessment of this tutoring system is based on the amount of training and the amount of positive and negative completed exercises of identified training categories. Principles from a cognitive architecture are used to keep close to aspects of human learning, namely frequency and recency of usage. Simulation runs demonstrate a successful adaptation of training: exercises of categories with which students have most difficulties are presented most.
Christian P. Janssen, Hedderik van Rijn