[REVIEW]: Recommendation.jl: A Framework for Building Recommender Systems in Julia

[editorialbot]

Submitting author: @takuti (Takuya Kitazawa)
Repository: https://github.com/takuti/Recommendation.jl
Branch with paper.md (empty if default branch):
Version:
Editor: @lucaferranti
Reviewers: @abhijithch, @PGimenez
Archive: Pending

Status

Status badge code:

HTML: <a href="https://proceedings.juliacon.org/papers/3b10d49b644c6be6fa3f78110a998783"><img src="https://proceedings.juliacon.org/papers/3b10d49b644c6be6fa3f78110a998783/status.svg"></a>
Markdown: [![status](https://proceedings.juliacon.org/papers/3b10d49b644c6be6fa3f78110a998783/status.svg)](https://proceedings.juliacon.org/papers/3b10d49b644c6be6fa3f78110a998783)

Reviewers and authors:

Please avoid lengthy details of difficulties in the review thread. Instead, please create a new issue in the target repository and link to those issues (especially acceptance-blockers) by leaving comments in the review thread below. (For completists: if the target issue tracker is also on GitHub, linking the review thread in the issue or vice versa will create corresponding breadcrumb trails in the link target.)

Reviewer instructions & questions

@lucaferranti & @abhijithch & @PGimenez, your review will be checklist based. Each of you will have a separate checklist that you should update when carrying out your review.
First of all you need to run this command in a separate comment to create the checklist:

@editorialbot generate my checklist

The reviewer guidelines are available here: https://joss.readthedocs.io/en/latest/reviewer_guidelines.html. Any questions/concerns please let @lucaferranti know.

✨ Please start on your review when you are able, and be sure to complete your review in the next six weeks, at the very latest ✨

Checklists

📝 Checklist for @PGimenez

[editorialbot]

Hello humans, I'm @editorialbot, a robot that can help you with some common editorial tasks.

For a list of things I can do to help you, just type:

@editorialbot commands

For example, to regenerate the paper pdf after making changes in the paper source files, type:

@editorialbot generate pdf

[editorialbot]

Software report:

github.com/AlDanial/cloc v 1.90  T=0.05 s (1757.8 files/s, 154268.9 lines/s)
-------------------------------------------------------------------------------
Language                     files          blank        comment           code
-------------------------------------------------------------------------------
TeX                             14            399            178           2938
Julia                           55            645            619           2306
Markdown                        12            144              0            303
YAML                             3              0              0             96
Ruby                             1              8              4             45
TOML                             2              5              0             23
make                             1              3              0              7
-------------------------------------------------------------------------------
SUM:                            88           1204            801           5718
-------------------------------------------------------------------------------

Commit count by author:

   330	Takuya Kitazawa
     2	Hieronimo
     1	Dhairya Gandhi
     1	Julia TagBot
     1	Tony Kelman

[editorialbot]

Paper file info:

📄 Wordcount for paper.tex is 146

🔴 Failed to discover a Statement of need section in paper

[editorialbot]

License info:

✅ License found: MIT License (Valid open source OSI approved license)

[editorialbot]

Reference check summary (note 'MISSING' DOIs are suggestions that need verification):

OK DOIs

- None

MISSING DOIs

- 10.1137/141000671 may be a valid DOI for title:  Julia: A Fresh Approach to Numerical Computing 
- 10.1145/3340531.3412778 may be a valid DOI for title:  LensKit for Python: Next-Generation Software for ...
- No DOI given, and none found for title:  MyMediaLite: A Free Recommender System Library 
- No DOI given, and none found for title:  LibRec: A Java Library for Recommender Systems 
- No DOI given, and none found for title:  The MovieLens Datasets: History and Context 
- 10.18653/v1/d19-1018 may be a valid DOI for title:  Justifying Recommendations using Distantly-Labele...
- No DOI given, and none found for title:  2nd Workshop on Information Heterogeneity and Fus...
- 10.1145/138859.138867 may be a valid DOI for title:  Using Collaborative Filtering to Weave an Informa...
- No DOI given, and none found for title:  Item-Based Collaborative Filtering Recommendation...
- 10.1145/3130348.3130372 may be a valid DOI for title:  An Algorithmic Framework for Performing Collabora...
- No DOI given, and none found for title:  Item-Based Top-N Recommendation Algorithms 
- No DOI given, and none found for title:  Application of Dimensionality Reduction in Recomm...
- No DOI given, and none found for title:  Matrix Factorization Techniques for Recommender S...
- No DOI given, and none found for title:  Netflix Update: Try This at Home 
- No DOI given, and none found for title:  Feature-Based Matrix Factorization 
- No DOI given, and none found for title:  BPR: Bayesian Personalized Ranking from Implicit ...
- No DOI given, and none found for title:  Multiverse Recommendation: N-Dimensional Tensor F...
- 10.1007/978-0-387-85820-3_3 may be a valid DOI for title:  Content-Based Recommender Systems: State of the A...
- 10.1145/2792838.2796542 may be a valid DOI for title:  Factorization Machines for Hybrid Recommendation ...
- 10.1145/2168752.2168771 may be a valid DOI for title:  Factorization Machines with libFM 
- 10.1145/2124295.2124313 may be a valid DOI for title:  Learning Recommender Systems with Adaptive Regula...
- No DOI given, and none found for title:  Social Network and Click-Through Prediction with ...
- No DOI given, and none found for title:  RecPack: An (Other) Experimentation Toolkit for T...
- 10.1007/978-0-387-85820-3_8 may be a valid DOI for title:  Evaluating Recommendation Systems 
- No DOI given, and none found for title:  A Survey of Serendipity in Recommender Systems 
- 10.1145/1060745.1060754 may be a valid DOI for title:  Improving Recommendation Lists through Topic Dive...
- No DOI given, and none found for title:  Julia 1.0 Programming Complete Reference Guide: D...
- No DOI given, and none found for title:  Julia Programming Projects: Learn Julia 1.x by Bu...
- No DOI given, and none found for title:  OFF-Set: One-Pass Factorization of Feature Sets f...
- No DOI given, and none found for title:  The Netflix Prize 
- No DOI given, and none found for title:  Introduction to Information Retrieval 
- 10.2139/ssrn.3378581 may be a valid DOI for title:  Recommender Systems and Their Ethical Challenges 

INVALID DOIs

- None

[editorialbot]

👉📄 Download article proof 📄 View article proof on GitHub 📄 👈

[lucaferranti]

@PGimenez, @abhijithch, thank you very much for volunteering as reviewers!

I will be the editor for this submission, feel free to ping me to ask any questions you may have.

You can find review guidelines here feel free to ask at any point if something is unclear.

As a first step, you can generate your checklist by running

@editorialbot generate my checklist

You can write your review comments here or directly open an issue in the paper repository

[lucaferranti]

@editorialbot remove @lucaferranti as reviewer

[editorialbot]

@lucaferranti removed from the reviewers list!

[lucaferranti]

Hi @abhijithch and @PGimenez 👋 ,

just checking in to see if you had time to start the review. Any timeline estimate?

[PGimenez]

Review checklist for @PGimenez

Conflict of interest

• I confirm that I have read the JuliaCon conflict of interest policy and that: I have no COIs with reviewing this work or that any perceived COIs have been waived by JCon for the purpose of this review.

Code of Conduct

• I confirm that I read and will adhere to the JuliaCon code of conduct.

General checks

• Repository: Is the source code for this software available at the https://github.com/takuti/Recommendation.jl?
• License: Does the repository contain a plain-text LICENSE or COPYING file with the contents of an OSI approved software license?
• Contribution and authorship: Has the submitting author (@takuti) made major contributions to the software? Does the full list of paper authors seem appropriate and complete?

Functionality

• Installation: Does installation proceed as outlined in the documentation?
• Functionality: Have the functional claims of the software been confirmed?
• Performance: If there are any performance claims of the software, have they been confirmed? (If there are no claims, please check off this item.)

Documentation

• A statement of need: Do the authors clearly state what problems the software is designed to solve and who the target audience is?
• Installation instructions: Is there a clearly-stated list of dependencies? Ideally these should be handled with an automated package management solution.
• Example usage: Do the authors include examples of how to use the software (ideally to solve real-world analysis problems).
• Functionality documentation: Is the core functionality of the software documented to a satisfactory level (e.g., API method documentation)?
• Automated tests: Are there automated tests or manual steps described so that the functionality of the software can be verified?
• Community guidelines: Are there clear guidelines for third parties wishing to 1) Contribute to the software 2) Report issues or problems with the software 3) Seek support

Paper format

• Authors: Does the paper.tex file include a list of authors with their affiliations?
• A statement of need: Does the paper have a section titled 'Statement of need' that clearly states what problems the software is designed to solve, who the target audience is, and its relation to other work?
• References: Do all archival references that should have a DOI list one (e.g., papers, datasets, software)?

Content

• Context: is the scientific context motivating the work correctly presented?
• Methodology: is the approach taken in the work justified, presented with enough details and reference to reproduce it?
• Results: are the results presented and compared to approaches with similar goals?

[PGimenez]

Hi, I'd like to submit my review of the paper "Recommendation.jl: A Framework for Building Recommender Systems in Julia" below.

I believe this is a good paper on recommender systems. It does a deep enough review on the most popular methods used in the literature, and provides implementation details in the package. The paper is well-written, with clear language and good usage of math notation to explain concepts. I think the pape is apt for publication with minor changes/corrections.

There are a few minor points I'd like to raise, which can also be found in the attached annotated PDF:

• It seems Figure 3 illustrates the fit and recommendation operations, not the data conversion into a DataAccessor

• is it possible to create a DataAccessor from a matrix R?
  R could indeed be converted into an event array by having the users and items be the row/column indices of the matrix, and the ratings be the entries of the matrix

• In 3.2.2 you mention that handling big matrices is very expensive. However, SVD is also known for being extremely costly ( O(N^3)) when used in matrix completion, as you mention in the next section 3.2.3. Therefore, SVD does not solve the resource cost problem.

• It is not clear to me how FMs relate to the (user,item) model used earlier, although I'll admit I hadn't seen FMs before.
  I assume x would be a feature vector for a user, and y(x) would yield a rating for this vector. But, how is x related to the item being rated?

• Just before 4.2.1, Sentence seems incomplete, I'd reorder them like:

let u be a target user, I the set of all items, and I_u the set of. truth items

Or leave as it is and add "be" before each variable like

let a target user be u, the set of all items be I...

---

Then, a few more points:

The cold-start problem

In 3.4 you state that the cold-start problem arises when "there is not enough historical data to capture meaningful information". The question then is, how much data is needed? I believe a single rating is enough.

To me, the issue here is adding new users or items to the system since these haven't rated or haven't been rated by anyone yet. Therefore the system cannot find items similar to the one the user likes (which is none since they've rated nothing)

From a formulation standpoint, new users and items have an empty row/column in matrix R. When factorizing R with any of the earlier methods, the coefficients associated with the new user/item (p_i, q_j for instance) will be all-zero. This would yield no recommendations.

This can be solved by leveraging additional information in the form of feature vectors (the user attribute parameter) for either users or items. In fact, I believe this is what the content-based filtering in 3.4 is actually doing since it requires the user preferences (but no prior ratings).

This usage of attribute vectors can be incorporated into the matrix factorization technique as well, as shown in these papers

Towards Open-World Recommendation: An Inductive Model-based Collaborative Filtering Approach

Matrix completion and extrapolation via kernel regression (section 3. Disclosure, I wrote this one 😄)

My point here is: If most of the implemented algorithms can't solve the cold-start problem, how usable are they in real-world situations?

Experimental results

The results are good for showcasing the different implemented metrics and having a better understanding of their meaning. However, I find it lacking that only the SVD method is used. I'd expect the experiments to also show how this package allows to easily switch between algorithms. Also, some benchmarks as to their computational cost.

Package documentation

The package documentation is good, as it covers the algorithms in detail. It'd be good to also have one or two tutorials like the script included in the paper. (I there are already, sorry, I couldn't find any)

[lucaferranti]

Thank you very much @PGimenez for the review!

@takuti please take a look at the comments and let us know when you have addressed the review comments

[PGimenez]

sorry, forgot to attach the annotated PDf =)
10.21105.jcon.00147.pdf

[takuti]

Thank you for the review comments, @PGimenez! Just informing you that I am actively working on the points and hopefully update you in the coming weeks. cc: @lucaferranti

[lucaferranti]

Hi @abhijithch 👋 ,

how is it going with the review? Do you happen to have an estimated timeline?

[lucaferranti]

Hi everyone 👋 , just checking in to see how this review is going.

@takuti did you manage to address the comments raised by @PGimenez .

@abhijithch how is it going with the review? Any expected timeline?

[lucaferranti]

@editorialbot assign @lucaferranti as editor

[editorialbot]

Assigned! @lucaferranti is now the editor

[takuti]

@lucaferranti Thank you for checking in and sorry for the late reply. I'm a bit occupied lately but making slow progress on addressing the review comments. I probably want to wait until I receive the second reviewer's comments so I can efficiently address all of them in a batch.

[lucaferranti]

Thanks @takuti, yes waiting for the second reviewer is reasonable.

@abhijithch do you have any updates on your review? If you are unable to complete the review, please let me know so that I can find another one

[abhijithch]

@lucaferranti, apologies for not replying sooner. I will work on this and will share the review comments soon.