Performance significantly superior to the original report

[ShugangZhang]

Hi authors,
Thanks for your great work! The reproduction is easy and friendly. I have one question here. In the original paper, the performance of GKT on ASSIST2009 is AUC=0.723 (https://dl.acm.org/doi/10.1145/3350546.3352513), however, running your code gives a much higher performance (AUC=0.820). I have also tried to seperate a "valid" set, but the result is still around 0.82. Could you please give any potential reason for such difference?
Thanks in advance!

Shugang

[hcnoh]

Hi. Thanks for your comment.

I think the dataset used in the GKT paper is not ASSIST2009 but ASSIST2015, because the performances of other baseline models look like that.

More details will be provided by asking the authors of the paper.