As an ML engineer working on an e-commerce recommendation system, you need to optimize for real-time learning and cost efficiency. The system must dynamically adapt to user preferences and seasonal trends while minimizing computational costs. Which approach best meets these requirements?