User-friendly guidance on data preparation

recommenders-team / recommenders

Best Practices on Recommendation Systems

https://recommenders-team.github.io/recommenders/intro.html

MIT License

18.89k stars 3.07k forks source link

User-friendly guidance on data preparation #375

Closed yueguoguo closed 5 years ago

yueguoguo commented 5 years ago

It takes time in real-world problems that user feel frustrated about preparing data for recommender algorithms. It would be good to have notebooks to detail that.

WessZumino commented 5 years ago

Totally agree! In the RBM deepdive I have used some space to explain the logic behind the data splitting and preparation for movielens. However, this is quite a simple scenario as the data are already given in the correct format. It would be good to find a real case scenario in which more work is needed for data preparation

ghost commented 5 years ago

May we please also have guidance for FFM dataset preparation to feed xDeepFM, especially with integrating customer and item features into the input dataset along with implicit/explicit feedback.

yueguoguo commented 5 years ago

@atimesastudios yea we do have some preparation functions for this

The negative_feedback_sampler (see here) is the one that samples negative feedback when data is about implicit feedback.
The LibffmConverter (see here) is the one that performs libffm format conversion from feature data set.