Regarding the issue with reproduction

[zhaijianyang]

Hello. I have encountered an issue with reproducing the results. I replicated two methods, Recformer and UniSRec, using default parameters. The results for Recformer seem reasonable, but the performance of the reproduced UniSRec is significantly better than reported in the paper. What could be the reason for this?

[JiachengLi1995]

Are you using UniSRec with text only or text+IDs? In this paper, we tested the text-only version of UniSRec.

[zhaijianyang]

I conducted experiments by incorporating IDs, and the results are as follows. What confuses me even more is that I did not pre-train either of the methods; I only utilized the pre-training parameters of the language model.

[JiachengLi1995]

Which category are you working on? Do you mean you didn't pretrain Recformer and UniSRec?

[zhaijianyang]

I'm sorry, I misspoke earlier. UniSRec actually performed worse after pretraining.