EMV-ADKVMN: An attention-based multi-vector knowledge tracing framework for educational innovation in adaptive learning systems

Abstract

The rapid advancement of deep learning is driving transformative changes in adaptive learning systems. In this work, we propose EMV-ADKVMN, a knowledge tracing framework that integrates both historical and real-time behavioral data. By synthesizing features at the input layer, our model extends the original Dynamic Knowledge Vector Memory Network (DKVMN) and achieves more effective knowledge state representation. Experiments on publicly available datasets demonstrate a mean AUC of 0.822, showing clear improvements over conventional methods. Beyond predictive performance, we implement our framework in an adaptive learning system. Results show that student satisfaction across different performance groups exceeds 80%, confirming the system’s adaptability and effectiveness. By enabling precise knowledge tracking and supporting personalized learning trajectories, our approach enhances educational outcomes and promotes sustainable use of educational resources. Overall, EMV-ADKVMN highlights a promising direction for future pedagogical innovation and sustainable educational practices.