search

NicolasHug / Surprise

A Python scikit for building and analyzing recommender systems
http://surpriselib.com
BSD 3-Clause "New" or "Revised" License
6.37k stars 1.01k forks source link

Slope One prediction formula does not make sense #384

Open mmlynarik opened 3 years ago

mmlynarik commented 3 years ago

Hi,

I'm a fan of your package, since it's very well documented and logically structured. Anyway, I wanted to bring up the topic of the meaningfulness of the Slope One formula. Are you sure the formula below actually makes sense, if the predicted rating is dependent exclusively on deviations evaluated based on ratings of all other users and just the average rating of given user? I mean, where is the contribution of the given user's preferences demonstrated by his ratings to the prediction of the new item rating? The only term which carries any kind of user-related information is his mean, which is quite negligible. You can check e.g. this document on page 34, which defines the Slope One formula in a more logical way, taking into account also user's ratings. http://guidetodatamining.com/assets/guideChapters/DataMining-ch3.pdf.

image

NicolasHug commented 3 years ago

Sorry I won't have time to dive deep into SlopeOne again, but you'll probably find some insights in the original paper https://arxiv.org/abs/cs/0702144

If the formula in the docs doesn't reflect the paper, please let met know!

mmlynarik commented 3 years ago

Sorry I won't have time to dive deep into SlopeOne again, but you'll probably find some insights in the original paper https://arxiv.org/abs/cs/0702144

If the formula in the docs doesn't reflect the paper, please let met know!

Thank you, I will check it and find out whether it solves my issue!