1
$\begingroup$

In the recommender system NCF, the input is a batch of user-item interactions (one-hot encoded) and the output is a 0-1 score of whether the item has been bought or not:

enter image description here

This seems to indicate that the item input vector that the model is trained on already contains y. I understand the purpose of this, but doesn't that lead to a dangerously high chance of overfitting?

$\endgroup$
1
+50
$\begingroup$

The input of the system is a given user and a given item (each respectively one-hot encoded). The output of the system is binary (picked or not).

The authors filtered the data to require each user to have picked at least 20 different items. Thus, enabling the system to learn the general preferences of a user.

The system could overfit. The system could memorize that users would only pick items they have already picked.

The system would fail on the chosen evaluation protocol - held-out last interaction. The model performs better than other recommendation systems on this evaluation protocol which is evidence that this system is not overfitting.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.