Both approaches you mentioned for creating the model are valid, depending on your goals:
Starting from the problem domain (conceptual schema) means you focus on the real-world concepts and design your Java classes to represent those directly. This approach keeps your model clean and focused on the business logic, which is great for maintainability and clarity.
Starting from the restructured database schema means your Java model closely reflects the database tables and columns, ignoring any extra surrogate keys you added. This can make data handling more straightforward but might couple your model tightly to the database structure.
In many projects, a mix of both is used: design your model based on the problem domain for clarity, then map it to the database schema via DAO implementations for data access. This helps keep the code organized and flexible.