Really nice! As an extension to the article, I'm also making a recommender, but just colab filtering. But yours looks stellar! And the article is great, compliments!
Need some time to let the like score calculation sink in :-)
I'm going to experiment with the (rating * 2) / 100, seems like a great way to account for the nonlinearity.
Btw don't you divide by 10 instead of 100?
Another suggestion was to take transform the ratings of each user to percentiles, as a measure of how favorite the game is to the user, also seems interesting.
Need some time to let the like score calculation sink in :-) I'm going to experiment with the (rating * 2) / 100, seems like a great way to account for the nonlinearity. Btw don't you divide by 10 instead of 100?
Another suggestion was to take transform the ratings of each user to percentiles, as a measure of how favorite the game is to the user, also seems interesting.