Data analysis for the detection of a learner blocking state as part of the QED-Tutrix intelligent tutorial system
A blocking state is a cognitive state in which a student cannot make any progress toward finding a solution to a problem. In this research, we present the development of probabilistic models to detect a blocking state while solving a Canadian high school-level problem in Euclidean geometry on an intelligent tutoring system. Our methodology includes an experimentation with a modified version of QED-Tutrix, an intelligent tutoring system, which was used to gather labelled datasets composed of sequences of mouse and keyboard actions. We observed blocking states in this dataset from subsequence distributions and frequency of states. Using a probabilistic framework, we developed four predicting models: an actionfrequency model, a subsequence-detection model, a 1D convolutional neural network model and an hybrid model. The hybrid model outperforms the others with a F1 score of 80.4 % on classification of blocking state on training set. It performs 77.3 % on test set. The applications of this research lead to an upgrade of QED-Tutrix internal finite-state machine for its interactions with the learner. Also, this research opens a second research stage, in which targeted tutorial interventions in QED-Tutrix can be developed. This can be achieved with an algorithm that understands the context of intervention and that is able to help precisely the learner. In order to get better performances from the current models, the history of the previous blocking states needs to be incorporated. Moreover, the mathematical concepts used by the learner can be integrated.
For more information, click here
Étude Prospective d’un Système Tutoriel à l’aide du Modèle des Espaces de Travail Mathématique
Proceedings Fifth ETM Symposium