Personalizing the Sequencing of Learning Activities by Using the Q-Learning and the Bayesian Knowledge Tracing

Abstract

In this paper, we present an approach for personalizing the sequencing of learning activities that relies on the Q-learning. The Q-learning learns a sequencing policy to select learning activities that aims to maximize the learning gain of students. On the one hand, the core of this approach is the use of the Bayesian knowledge tracing (BKT) to model the student knowledge state and to define the Q-Learning reward function. On the other hand, we defined with experts rules to generate simulated students. These simulated data were used to initialize the Q-table of the Q-Learning and answer its“cold start” problem. We present empirical results showing that the sequencing policy learned from the expert-based initialization of the Q-table provides the system with an efficient strategy to improve the students’ knowledge states in comparaison with the Q-table randomly initialized. We further show that Q-Learning approach based on the knowledge states of the students inferred by the BKT are promising way for adaptive instruction in intelligent tutoring systems.