Dynamic ensemble learning has become a pivotal paradigm in the evolution of recommendation systems (RSs), enabling adaptive, context-aware, and robust predictive performance through the real-time integration of diverse learning models. While ensemble-based approaches have been extensively studied, prior review articles have predominantly focused on static frameworks and often lack a systematic synthesis of dynamic ensemble strategies. More critically, these reviews provide limited insight into methodological challenges and do not sufficiently outline future research trajectories necessary for advancing the field. This paper presents a comprehensive and structured review of dynamic ensemble learning techniques within RSs, encompassing theoretical underpinnings, architectural classifications, learning mechanisms, and model selection strategies. Emphasis is placed on recent advancements incorporating deep neural networks, reinforcement learning, and context-adaptive ensemble architectures. Furthermore, the review critically examines key challenges, including data stream volatility, computational scalability, model interpretability, and the integration of user-centric contextual signals. By consolidating current developments and identifying unresolved research questions, this review offers a forward-looking perspective aimed at guiding future exploration and innovation in dynamic ensemble-based RSs.
Sharma et al. (Sat,) studied this question.