Hacker Newsnew | past | comments | ask | show | jobs | submit | amplifier_khan's commentslogin

Aside from algorithmic gender bias and "mathwashing" as mentioned in the article, what other sources need be considered?


Yup it seems like currently recs are based on content filtering techniques, based on defined properties of each beer. I'd be interested in seeing this project utilize a collaborative filter to take into account user interactions/ratings associated with each beer and then predict beers that people would enjoy based on their preferences.


I used collaborative filtering for movie recommendation (http://web.onetel.com/~hibou/morse/MORSE-Paper.html), and it worked well for that, and I'd probably use it for music too. Movies and music have quite complex "structure", making it hard to compare even similar examples. Dr Strangelove's most similar movie is probably Fail-Safe, but the two are still very different in many ways, aren't really interchangeable, and I wouldn't be surprised if someone liked one but not the other.

For beer, although there are many different styles, similar beers could more easily be substituted for each other (e.g. Sierrra Nevada Pale Ale and Brewdog Punk IPA). So if you couldn't get the one you asked for, you'd be happy with the other.

I have no hard evidence, though, and might be wrong. I didn't get the chance to apply my MORSE algorithm to beer, and it wasn't as generally applicable as I'd have liked. (See http://web.onetel.com/~hibou/morse/DeadOnTime.html)


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: