Architectural Design and System-Level Evaluation of a Modular Adaptive Learning Recommender in Higher Education

Abstract

Many implementations of adaptive learning systems increasingly use data-driven personalisation, but they may be constrained by tightly integrated architectures that limit scalability, evolution, and adoption. This research reports on the system-level architectural and design evaluation of a modular adaptive learning recommender system developed as an extension to a production Moodle LMS. By adopting the design science research methodology, the modular content-enhanced collaborative filtering system provides a decoupling of data ingestion, recommendation computation, and model training via loosely coupled and standardised interfaces. A multi-level mixed-methods evaluation that combined system analytics with perception data from 1,051 learners and 119 e-learning professionals was conducted. The results show high accuracy in recommendations, efficiency in retraining models, and positive perceptions towards scalability, robustness, maintainability, and organisational adoption. In contrast to proposing a new algorithm, the research offers evidence related to the architecture and assessment of sustainable adaptive learning systems within an LMS environment.