YuhanZhen
commented
1 year ago Thank you for your questions. I greatly apprecate pointing out the typo errors for both our work and SR-GNN. Yes, you are right, the correct evaluation metric is MRR not MMR. I will correct it right away. As for the second question, in my perspective, the precision, recall and hit rate are the different evaluation metrics in mathmatics. However, in many papers with regard to the next item recommendation, we do not strictly distinguish them. That is to say, for the next one item recommendation, they express the same meaning actually. I am so sorry to confused you. Hopefully, I could answer your questions.
xxijian
Dear author,
My research direction is session-based recommendation. After reading your paper, I have a few small questions that I hope you can help me answer. Thank you so much in advance:
In the code, the evaluation metric is MRR, but it is output as "MMR." This is an output error in both the SR-GNN and GCE-GNN codes, and should be corrected to MRR, correct?
In the GCE-GNN code, after calculating the average of the hit rate, it is multiplied by 100, which should represent HR (Hit Recall), right? However, the model's print output writes "Recall". Additionally, in the papers for both SR-GNN and GCE-GNN, the evaluation metric used is P (precision), while in your paper, the evaluation metric used is HR. I would like to know how you obtained your HR metric and whether it was not written in the code. Do you know how they derived the P metric in their papers?
I'm really puzzled by these metrics, and have been struggling with them for a long time. Your paper has been very helpful to me in my work, and I would greatly appreciate it if you could help me answer these questions. Thank you very much!