You are developing a recommendation system to suggest products based on users' past purchases and browsing history. The initial model performs well on training data but poorly on unseen test data, indicating overfitting. To improve model scalability and generalization on new data, which strategy would be MOST effective?