Could you provide code for drawing figures (explain popularity bias).

[ChengHSUHSU]

for example, figure1. thanks a lot.

[ChengHSUHSU]

Hi, why do we regard y_ui as ground truth for predicting popularity item score??

[weitianxin]

Thank you for your inquiry. We optimize the objective to capture the direct relevance between the item and the recommendation. To achieve this, we have developed a branch that only inputs the item embedding to predict the recommendation targets ($y_{ui}$). By doing so, the bias between items and recommendations can be measured. Then during the inference stage, we use this measurement to remove such bias, as described in our paper. Of course, there are other potential methods (e.g. self-supervised learning) that can be used to capture the bias/relevance.

[489597448]

for example, figure1. thanks a lot.

Have you found the drawing code?