Developing a Theoretical Framework for Context-Aware Recommender Systems in Low-Bandwidth Nigerian Mobile Environments

Recommender systems operating in low-bandwidth mobile environments require theoretical reconceptualisation that accounts for intermittent connectivity, limited on-device processing, and contextual factors specific to Nigerian usage patterns. This study develops an original theoretical framework for context-aware recommender systems adapted to low-bandwidth Nigerian mobile environments. The framework extends the Contextual Pre-Filtering Model of Adomavicius and Tuzhilin by integrating three Nigeria-specific context dimensions: bandwidth availability state, language code-switching patterns, and temporal usage clustering attributable to shared device ownership. A prototype system was implemented using collaborative filtering with contextual post-filtering and evaluated against baseline recommender algorithms using a dataset of 180,000 user interactions collected from a Nigerian agricultural information mobile application. Evaluation metrics included precision, recall, novelty, and recommendation delivery success rate under simulated variable bandwidth conditions. Available recommender systems literature from low-resource African contexts identifies offline-first recommendation caching and lightweight matrix factorisation as the most technically effective approaches for bandwidth-constrained environments. The Activity Theory and Contextual Integrity Theory provide additional theoretical grounding. Experimental results demonstrate that the proposed framework achieves 23 percent higher recommendation delivery success rate than standard CF approaches under bandwidths below 500 kbps. The study's original contribution is a theoretically grounded, empirically validated recommender framework specifically calibrated to Nigerian mobile contextual constraints. Keywords: recommender systems, low-bandwidth, context-aware, Nigeria, mobile applications.