Hi,
I noticed that, during validation and testing, the model will be updated too. So, when comparing to other model mentioned in your paper, have you guys conduct the some configuration(update model parameters) during validation and testing?
Anyway, in my opinion, while each interaction in validation and test is only seen once, the involved user/item appears multi times, such that updating models in these two procedures is [data leakage ] to some extent. (Due to the existing of static one-hot embedding, updating model will update the corresponds weight parameter of the RNN cell.)
Hi,
I noticed that, during validation and testing, the model will be updated too. So, when comparing to other model mentioned in your paper, have you guys conduct the some configuration(update model parameters) during validation and testing?
Anyway, in my opinion, while each interaction in validation and test is only seen once, the involved user/item appears multi times, such that updating models in these two procedures is [data leakage ] to some extent. (Due to the existing of static one-hot embedding, updating model will update the corresponds weight parameter of the RNN cell.)