Studying Memory Decay and Spacing within Knowledge Tracing

Abstract

Knowledge Tracing estimates a student’s knowledge on a set of skills and predicts whether the student will answer correctly if given a question linked to subsets of such skills. We conduct an in-depth analysis on the best ways to apply the cognitive science principles such as memory decay and spacing within knowledge tracing, proposing new algorithms, MemDec and MemDec Spacing, to do so. We explore different methods of modeling the rate and weight of decay, with and without spacing, and analyze their impact on predicting performance in real-world data. Variations of the model are compared between each other as well as against other existing algorithms.