Quick question about the Inconsistent between Diffnet code and paper

[Dingseewhole]

In your great original paper Sec 3.2 you use a sigmoid-based BPR loss (i.e., Eq(13)) to train the model. And in Eq(13) you use (r_ai-r_aj). But in your implement code diffnet.py line-127 you use self.opt_loss = tf.nn.l2_loss(self.labels_input - self.prediction) to generate the loss. It means you use the MSE-loss which is greatly different from BPR loss. I wonder to know which one is the correct version?

[PeiJieSun]

Hi, thank you for your attention to our work. Please refer to this github version, it is the correct one. And sorry for the wrong loss function in the paper.

[Dingseewhole]

Thank you for your quick reply!
In Diffnet and Diffnet++, it seems you use the implicit dataset (such as Yelp, larger than 3 is pos others are abandoned) to train and test the model. And as I know the MSE-based loss is usually used in the explicit dataset that has both positive and negative labels. So if the loss function of Diffnet is MSE-based loss how did it get the negative inputs?

[PeiJieSun]

Sorry for the later response. You can treat the label of the negative inputs as zeros.