Afsana Nasrin Master’s Thesis Defense, Tuesday, November 18, 2025 @ 11:00 am Central Time

COMMITTEE CHAIR: Dr. Xishuang Dong

TITLE: ENHANCING LEARNING PATH RECOMMENDATION VIA MULTI-TASK LEARNING

ABSTRACT: Personalized learning is a student-centered educational approach that acknowledges and adapts to the unique characteristics of each learner by customizing content, pacing, and assessments to meet individual needs. Key components include learner profiles, cognitive styles, learning objectives, adjustable pacing, and enabling technologies such as intelligent learning environments, intelligent tutoring systems, and data mining. A key strategy for implementing personalized learning is learning path recommendation, which guides learners through tailored sequences of content. This approach sequentially recommends personalized learning items (e.g., lectures, exercises) to address each student’s specific needs. Recent advances in deep learning, particularly deep reinforcement learning, have made learning path recommendations more practical and expressive. However, most of the existing methods generally focus solely on learning path recommendations without fully considering the contributions of related tasks, such as knowledge tracing. This thesis proposes a multi-task learning model to enhance learning path recommendation by leveraging shared information across multiple related tasks. The approach reframes learning path recommendation as a sequence-to-sequence (Seq2Seq) prediction problem, generating a personalized learning path based on a learner’s historical interactions. The model employs multi-task deep learning to jointly optimize two tasks: learning path recommendation and deep knowledge tracing. Its architecture consists of a shared Long Short-Term Memory (LSTM) layer to capture common features and two task-specific LSTM layers to model the unique objectives of each task. To mitigate redundant recommendations, a non-repeat loss is incorporated to penalize repeated items in the generated learning path. Extensive experiments on the ASSIST09 dataset demonstrate that the proposed model substantially improves the effectiveness of learning path recommendations.

Room Location: Electrical Engineering Conference Room 315D