Recommenders: [FEATURE] Consider adding BPR model with a tutorial

[tqtg]

Description

Bayesian Personalized Ranking (BPR) is a well-established model for implicit feedback. I would like to contribute a tutorial using BPR model with the usage of Cornac library @ https://github.com/PreferredAI/cornac

Expected behavior with the suggested feature

Other Comments

[gramhagen]

Definitely agree. Would love to see your contribution. Putting it as a notebook in the notebooks/02_model section would make sense. You can add that as a dependency to the conda environment to support this.

[miguelgfierro]

hi @tqtg, it would be nice to have your contribution. Ideally, what we are doing with other algos is to use our dataloaders and splitters.

Then with that data we will put it into your model. Normally we will use model.fit but it can be any other method your library has. Then we will have either model.predict for rating or model.recommend_k_items for ranking. In this part, if the library is already developed, we normally try to add the minimum possible addition. Maybe a good example is the surprise utilities. Here the functions that we have are a small refactoring of the original content of surprise library, but we don't reinvent the wheel here.

And then finally, we will input the results on our ranking or rating metrics.

We have two kinds of algorithm content, one are the quick starts, where we just show quickly how to run an algo. The others are the deep dives, where we explain the math and show an algo implementation.

I think for your case it would be good to do a deep dive.

Please let us know if you need any help in designing the content. We would be happy to help

[tqtg]

Thank you for your instructions @miguelgfierro !
I'm working on my PR and will seek your suggestions when I have the first draft.

[tqtg]

Since we already have BPR model, I would like to add it to the benchmark notebook. Is it something we are interested in?

[gramhagen]

yes, definitely. I'll add a separate issue so we can close this out.