Course talk:CPSC522/Restricted Boltzmann Machines for Collaborative Filtering

From UBC Wiki

Contents

Thread titleRepliesLast modified
Peer Reviews118:49, 28 March 2019
Peer review023:24, 17 March 2019

Peer Reviews

I clearly remember commenting on your page in here on March 20, but I don't know why it does not exist right now! I had mentioned that you had not written your thoughts on the incremental contribution of the papers, which I see you have added that since March 20. I had also mentioned that I understand how the second paper is related to the first paper, but the text in the conclusion section leads the readers to feel like the two papers are only related in solving the same problem. Best, Ali

AliMohammadMehr (talk)06:53, 28 March 2019

Hi Ali, thank you for checking back and leaving the feedback.

> the text in the conclusion section leads the readers to feel like the two papers are only related in solving the same problem.

RBM at the time of publication was (I believe) a de novo approach to collaborative filtering, so RBM and SVD methodologically are not very closely related. I'll add some details to avoid confusion.

NamHeeKim (talk)18:49, 28 March 2019
 

Peer review

(I am not very familiar with machine learning, so that I might say something very stupid.)

First of all, as usual, your writing is very clear, which I like.

I suppose that the two approaches you discussed in the entry are not related but address the same research problem (scalable recommender system, roughly speaking). Did Salakhutdinov et al got motivated by some perceived limitation of Sarwar et al? I guess that the former said something in the previous works section about SVD in "advertising" their own approach?

Also, you mention some difference in the error patterns in the evaluation section. Maybe it is not relevant in establishing the advantage of RBM, but you could elaborate it a little -- I am just curiuous of the cause.

ShunsukeIshige (talk)23:24, 17 March 2019