EARS: A federated and explainable recommendation system approach

Recommendation systems are increasingly central to digital platforms, yet they continue to face persistent challenges involving privacy, data scarcity, transparency, and cross-domain adaptability. The Environment Adaptive Recommendation System (EARS) addresses these challenges through a federated architecture that enables cross-domain collaborative learning while retaining raw interaction data locally. This framework integrates Federated Learning (FL), Hybrid Recommendation Algorithms, and explainable Artificial Intelligence (XAI). Three complementary component-level studies examined hybrid prediction, federated-training feasibility, and local interpretability. The results highlight three findings: (1) the ALS + Embeddings configuration achieved an RMSE of 0.263 and precision of 0.988 in the simulated cold-start setting; (2) federated training incorporating a related auxiliary domain achieved a lower MSE than the Fashion-only centralized baseline (0.036 vs 0.053); and (3) in a separate Random Forest explanation study, SHAP captured interaction-feature contributions more consistently than LIME across the examined scenarios. Together, these studies support the development of EARS as an integrated architectural framework for collaborative recommendation with local data retention and instance-level explanations.