Why would you choose a model that needs training to update your score?

[BrickDanD]

By reading your paper, I think score updating can be achieved through simple functions, but why did you choose to use in-depth learning? And I don't understand what role it plays in the score update function. Can you help me answer my question?