Extracting Knowledge in a Game-Based Learning Environment From Interaction Traces

organization of the pedagogical domain knowledge. In this environment, the domain expert is not necessarily aware of the target audiences' knowledge levels. Consequently, there could be a gap between what the domain expert thinks is the right way to organize the domain knowledge and how the domain knowledge should be organized to maximize the learners' learning. In this context, we present a novel approach to fill this gap by the semi-automatic reorganization of the domain knowledge in a way that can potentially maximize students' learning. Our work is in the context of the serious game "Tom O'Connor." This game is designed as part of project CLES to rehabilitate/test different cognitive abilities among children with some cognitive disabilities. In order to provide adaptive learning to different learners, we have developed a platform GOALS (Generator of Adaptive Learning Scenarios) that records the learners' activities in the form of interaction traces and use them as knowledge sources to generate pedagogical scenario. A pedagogical scenario is a suite of game-based activities presented to the learner in order to achieve one or more pedagogical objectives. In this paper, we present a novel approach for updating knowledge domain and learner profiles from the interaction traces. The results of the updating process are then presented to the do-main expert who can approve or disapprove them accordingly. We will look for two kinds of update, namely: 1) the detection of new concepts in the domain model, 2) the detection of new links between the domain concepts and the pedagogical resources. We apply mining algorithms to classify different students, according to their responses and then perform the analysis. We present our approach's formalization and some validations.