Course talk:CPSC522/Multinomial Variational Autoencoders for Predicting Gender in MovieLens

Overall a good project, each component is explained clearly and the objective is clear. I agree with Sarah's opinion that gender analysis seems to need more motivation. I appreciate the VAE justifications as well as the derivation. The approach to these hypotheses is very nice and the table is clear. Some of the sections can be ordered a bit differently as rn a lot of the sections are unstructured and doesn't have a nice logical flow. I would break it into more sections with subsections. minor edits:

• an user's -> a user's
• hypothesis -> hypotheses

I thought the article was well-written and the details to present of the multinomial VAE were well chosen! Each hypothesis and the results were clear and it was easy to compare performance using the table!

I am wondering if there is a specific motivation for predicting the gender of users.

Also, in the conclusion, the reason for the results on the new binarization method is discussed. It would be interesting to see the average number of ratings for the users whose gender was predicted correctly compared to those users whose gender was not predicted correctly.

Minor errors

Abstract

• take in an user's -> take in a user's

Critique

1) Overall, this is a very well-written project report. I especially appreciate the justification for all main choices that were made. The formalism is easy to understand and it is an appropriate length for a final project report.

2) The "challenges" section is great. It would be even better if you can summarize how you address these challenges in the latter sections so that reader has some context on what the solution would look like.

3) Under the "Does changing the binarization method improve predictions" section, the content can be structured better. First explain why you selected the binarization method that you did, and then follow it with what would make more sense instead. For example:

"Previously, we stated that the ratings matrix is binarized such that an 1 indicates that the user gave the movie a rating of 4 or higher. The reason for the existing binarization method from Liang et al. comes from recommender systems. Since we want to recommend good movies, we should not give the same value of 1 to both good and bad movies. In contrast to this approach, we now propose a second binarization method. Here, an 1 indicates that the ..."

Minor grammatical errors

1) Likewise, for Amazon and other e-commerce platforms, the system would take *a user's purchases and their...

2) Here, *a 1 indicates that the user rated the movie.