Dynamic Learner Profiling and Engagement-driven Adaptation Using Large Language Models
Year: 2026
Type: article
Abstract: Dynamic learner profiling has become a critical component in adaptive learning systems, particularly with the emergence of large language models (LLMs) capable of processing complex natural language interactions. This paper presents a framework for engagement-driven adaptation using GPT-based large language models, behavioural analytics, and reinforcement learning to continuously update learner profiles in real time. The study adopted a design science and experimental research approach in developing a multi-dimensional learner profiling system that models cognitive, affective, behavioural, and metacognitive learner states. A Double Deep Q-Network (DDQN) was integrated to optimise instructional adaptation strategies, while retrieval-augmented generation improved contextual response generation. Experimental evaluation was conducted using public educational datasets and a proprietary tutoring interaction corpus containing over 1.8 million interaction records. The proposed framework was compared with static rule-based systems, vanilla GPT models, and profiling systems without reinforcement learning. Results showed significant improvements in profiling accuracy (overall F1-score of 0.86), learner engagement (65% deep engagement rate), knowledge retention (84%), and learning efficiency (1.5× improvement) compared with baseline systems. The findings demonstrate the potential of large language models and reinforcement learning to support highly personalised and adaptive educational systems capable of improving learner engagement, retention, and overall learning outcomes.
Source: FUDMA Journal of Sciences
Institution: Babcock University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
SDG: Quality education
Open Access status: diamond
OpenAlex ID: W7162092001