Lower Than Expected Scores When Reproducing H&M Recommendation Task Results

[DanialNejad]

I have been trying to reproduce the results of the recommendation task on the H&M dataset as outlined in the associated paper, but so far, I have been getting significantly lower scores than reported. I have followed the instructions and steps provided in the documentation and paper, using the specified hyperparameters and dataset.

Steps Taken:
git clone https://github.com/snap-stanford/relbench
cd relbench/examples
python idgnn_link.py --dataset rel-hm --task user-item-purchase

Despite following these steps, the scores I’m getting are consistently lower than the ones reported in the paper. I suspect that there might be additional details or steps not mentioned explicitly in the paper that could be affecting the results.