Abstract
<title>Abstract</title> <p>In most existing knowledge tracing models, students' knowledge state is assumed to constantly change with the variation of individual exercises and even allowed for significant fluctuation. However, students usually work on exercises in batches within short time windows, separated by relatively longer intervals during which forgetting and latent learning may occur. Such special application scenarios require optimized knowledge tracing models to support the tracing of students' knowledge state with characteristics of stability in short time range and variability across different time stages. In this paper, a knowledge tracing model is developed based on knowledge-context-aware augmentation of Q-matrices for the tracing of students' intermittent knowledge states. Particularly, an approach of topologically sorting knowledge concepts is proposed by exploiting the complete paths contained in the knowledge concept graph. Then, a method of augmenting original Q-matrices is developed by requiring the updated Q-matrices to have best compatibility with the topological sorting on knowledge concepts. Finally, based on the augmentation of Q-matrices, a knowledge tracing model is developed by exploring Bayesian formulas and item response theory, which is specifically designed to trace students' intermittent mastery on knowledge concepts. Data experiments are conducted on three public datasets and two newly collected datasets in order to verify the efficiency of the proposed model.</p>