Integrating Working Memory into Learning and Content Analytics for Personalized Learning Systems

Personalized learning systems use learning and content analytics to trace student concept knowledge over time. We propose Trace-Memory, a machine-learningbased framework for learning and content analytics that incorporates concepts from cognitive load theory. A key limitation of existing learning algorithms is that student response models oversimplify test performance as a function of only the inherent difficulty of the question and the students mastery of the concepts underlying each question. However, research in cognitive science has shown that students performance can also be affected by their working memory capacity and the cognitive load of the question. In this work, we develop a new dynamic student response model that measures students working memory capacity and each questions cognitive load, and includes these terms in the student response model. Preliminary experimental results using the proposed model on synthetic and real-world data have shown promise, but also suggest that further improvements are needed to improve the models sensitivity to noise. Ultimately, this research has the potential to result in personalized learning systems that can identify students with low working memory capacity and improve their educational experience by customizing learning resource presentation to better cater to their needs.