Course talk:CPSC522/Matrix Factorization For Recommender Systems

The papers need to be in the references at the end. (How is the Bibliography annotated?)

Each Figure need an explicit reference on the page. You cannot take a figure from somewhere and use it as you own. (Fair use is okay if you refer to the content in your text, but you need to give explicit reference so no one will think it is your own work.)

In the Matrix definition, keep the upper-lower case distinctions clear (I think one j should be J). Tell us that I is the number of customers and J the number of movies.

Minimizing squared error (on a held-out test set) was the goal of the Netflix prize; it should be factored out of the Approach 1. The regularization you state in Approach 2 is also used in Approach 1; these are different algorithms to perform the same optimization.

Please use the notation consistently; how are q and p related to u and m?

I'm not sure that the modification in "Modifying MF Calculations" are modifications. Isn't this equivalent to fixing one row/column of u and m to be 1?

Which of the papers was the "implicit feedback" from? Why/how does |N(u)|^{0.5} normalize this value?

"MF can beautifully consider these effects" seems like a strange comment. How does p_ depend on time? One of the problems is that most users have very few data points; estimating temporal changes seems even more difficult. Why are items static? Surely the interest in a movie changes over time!

I don't understand the confidence. Where does it come from? How can it be observed or learned? (Is c_ui a number multiplied or a function (your notation is ambiguous).

In the plot, what is v1 and v2?

Thanks for your comments. I addressed those issues. To clarify: 1. Modifying MF Calculations is a section that has subsections: Adding Biases, Additional Input Sources, Temporal Dynamics, Inputs with varying confidence levels. These are not applicable by just fixing one row/column of u and m to be 1. 2. Implicit feedback is from second paper. As you can read in the page, all these modifications are from second paper. 3. Unfortunately, the authors does not explain a lot about v1 and v2. So, this is not that obvious for me too. But, I added a sentence from the paper about that.

I found your article very interesting. I also found the math behind the models easy to follow which was good.

The first piece of feedback I have is about the adding biases paragraph. I don't understand the sentence "the average rating is 3.5 but this movie is better than average". How does the model decide which movies are actually their average rating vs which movies deserve a higher average rating?

The second piece of feedback I have is to make sure you cite the two papers you are referencing in the body of your paper. This would help the reader keep track of which models come from which paper and how the two papers are connected to each other.

Overall I thought your paper was well-written and easy to follow!

In this evaluation, it is essential to address several critical points to enhance the clarity, accuracy, and completeness of the discussed content. First and foremost, the papers under discussion must be properly cited in the references section at the end of the text. This is a fundamental scholarly practice that ensures due credit is given to the original authors and facilitates further reading for interested individuals. Furthermore, it is imperative to explicitly reference any figures used in the text to avoid any potential misinterpretation of their origin. By doing so, readers can clearly discern the source of the figures and differentiate between original content and external references.

Additionally, maintaining consistent notation is paramount in clarifying the relationships between variables in matrix factorization explanations. Specifically, the interplay between 'q,' 'p,' 'u,' and 'm' should be clearly defined. Furthermore, any modifications in the "Modifying MF Calculations" section should be distinctly differentiated from the practice of merely fixing one row or column of matrices. Addressing these concerns will result in a more precise and reader-friendly discussion of matrix factorization for recommender systems. Finally, the article should provide more explicit and thorough explanations for several key concepts, such as "adding biases," "implicit feedback," "temporal dynamics," and "confidence levels." These terms require a more comprehensive introduction to ensure readers fully understand their relevance and implications in the context of recommender systems. By addressing these points, the article will offer a more complete and informative discussion of the topic.