Re:Netflix Replacing Star Ratings With Thumbs Ups and Thumbs...
by StefanKarpinski on Friday March 17, 2017 @03:11PM
(#13897020)
I've built recommender systems at Etsy, among other places, and you get much better results if you make the data yes/no first. Shockingly MUCH better results. You'd think it would help to take into account, say how _many_ items of a shop a user likes, rather than just the fact that they liked some item, but it doesn't – it actively hurts. The only meaningful piece of information, it turns out is does this person like this thing or not.
[ Parent
| on HN ]
Re:Netflix Replacing Star Ratings With Thumbs Ups and Thumbs...
by ghaff on Friday March 17, 2017 @03:25PM
(#13897149)
That doesn't really surprise me. A lot of people imagine that highly granular, multi-dimensional rating systems must be best and they're often not. Of course, it depends a bit on what you're trying to distinguish but then it may not really matter to distinguish among the 75% that people are genuinely happy with.