Why does loss_r_test become larger and larger when pre-training? about idcf HOT 3 CLOSED

Thanks for answering my question! I want to do some research based on your model and I also have the following questions: 1. There's a formula output = (ratings + x3) / 2. + user_b + item_b in your code. However, user_b for test users are not trained so it is meaningless. Could I remove user_b if I do not require more accuracy for trained users? 2. Are there any other terms that I can remove if I am willing to sacrifice some performance? 在2022-07-18 14:28:59，Qitian ***@***.***写道： Hi, the pretraining stage aims to obtain embeddings for key users that will used for computing inductive representations for query users in the next stage. Therefore, it is normal that the pretrained embeddings for testing are not accurate. For more details about two-stage training and the inductive CF models, please look at our paper. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>