Probabilistic Approaches to Detect Blocking States in Intelligent Tutoring System

Abstract

A blocking state is a measurable state on an intelligent tutoring systems’ user interface, which mirrors a student’s cognitive state where she/he cannot temporarily make any progress toward finding a solution to a problem. In this paper, we present the development of four probabilistic models to detect a blocking state of students while they are solving a Canadian high school-level problem in Euclidean geometry on an ITS. Our methodology includes experimentation with a modified version of QED-Tutrix, an ITS, which we used to gather labelled datasets composed of sequences of mouse and keyboard actions. We developed four predicting models: an action-frequency model, a subsequence-detection model, a 1D convolutional neural network model and a hybrid model. The hybrid model outperforms the others with a $F_1$ score of 80.4% on the classification of blocking state on validation set while performing 77.3% on the test set.