COMMITTEE CHAIR: Dr. Xiangfang Li
TITLE: GENERATIVE ARTIFICIAL INTELLIGENCE ENHANCED DEEP KNOWLEDGE TRACING FOR PERSONALIZED LEARNING
ABSTRACT: In today’s educational landscape, the demand for personalized learning experiences has gained significant attention, driven by advances in Artificial Intelligence (AI) and deep learning technologies. This dissertation investigates the integration of Generative Artificial Intelligence (GAI) with Deep Knowledge Tracing (DKT) to advance Personalized Adaptive Learning (PAL) systems, particularly within Historically Black Colleges and Universities (HBCUs). While HBCUs play a pivotal role in expanding educational opportunities, they often face challenges such as lower retention and graduation rates compared to other institutions. This research begins by exploring the theoretical foundations of personalized learning and DKT, a data-driven technique that models learner knowledge acquisition over time. Using four years of educational data (Fall 2020 to Summer 2023) from Prairie View A&M University (PVAMU), this study aims to enhance STEM education by predicting student course outcomes and identifying at-risk students. Multiple state-of-the-art (SOTA) DKT models, including DKT, DKT+, DKVMN, SAKT, and KQN, are employed to evaluate knowledge tracing performance. Experimental results reveal that SAKT and KQN consistently achieve the superior predictive accuracy, AUC, and F1 scores, enabling faculty and advisors to proactively support students through timely interventions. A key advancement of this study is addressing the challenges of data scarcity, which often limits the effectiveness of DKT in resource-constrained environments like HBCUs. To overcome this, GAI models such as TABSYN, TabDDPM and GReaT are utilized to generate synthetic datasets that augment real student records. The integration of tabular GAI enhances the robustness of DKT models, resulting in improved prediction accuracy and expanding the applicability of PAL systems across diverse educational contexts. By demonstrating the efficacy of synthetic data in strengthening DKT frameworks, this work highlights the potential of GAI to enhance the effectiveness and accessibility of PAL technologies. In conclusion, this dissertation advances the field of DKT by integrating innovative approaches that enhance PAL systems at HBCUs. It demonstrates how combining DKT models with GAI for synthetic data augmentation can significantly improve educational outcomes. Furthermore, it highlights the critical collaboration between AI researchers and educators in developing data-driven techniques that empower institutions to better support their students. By enhancing resource allocation, enabling proactive interventions, and refining support strategies, this research ultimately contributes to improving student retention and graduation rates.
Room Location: ECE 315D Conference Room