What is the benefit of having two separate class one for user and one for query?

[abdollahpouri]

Hello

I see in this tutorial, there are separate classes for user model and query model. The same for the movie model and candidate model. Can you please explain what is the rationale? In other words, why is it important to do that, and in what situations? Why not just one class for the Query (including user info) and one class for the Candidate?

[rlcauvin]

I don't think it's necessary to have separate classes for the user and query model or for the movie and candidate model.

The tutorial is showing how to add a deep neural network (DNN) to the existing towers. You could do the same thing by having a single Query class that handles the user and context feature embeddings and then runs the embedding outputs through a neural network, and a single Candidate class that handles the candidate embeddings and runs their outputs through a neural network.