Enhancing Attention-Based Knowledge Tracing with Digital Textbook Interaction

Abstract

Knowledge Tracing (KT) models a learner’s knowledge state by analyzing past responses to predict future performance. While traditional KT models focus on response correctness, few studies incorporate learning activity data during study sessions. This study proposes a KT model that integrates features from digital textbook viewing logs to enhance knowledge estimation. Experiments on a university course dataset demonstrate that incorporating study-related contextual information improves prediction performance, highlighting the impact of digital learning behavior on KT.