My cofounder and I thought about entering this market a year or two ago. If anyone cracks it open, there's a big prize at the end. TrunkClub et al are terrible at recommendations, they just push you the TrunkClub style and you either like it or you don't.
My concern with the automated style-learning approach is that, if Amazon still can't show me anything I want to buy successfully, then it must be a *really* hard problem. I can't think of any startup that's tackled recommendations successfully, though Pandora is probably closest.
--
Tried the app but it's pretty buggy for me right now. Logged in with Facebook, then nothing happened. Then I tried browsing as a logged out user anyway, and no items load within each category. Bummer. Will try again later.
@willimholte Awesome! Why didn't it get fully realized? Can you tell me more about the algorithm? (Was the theory basically: the more your friends like a place, the more you'll like it as well?)
@staringispolite I can't say for sure why they stopped working, it was a bootstrapped project that had typical startup growing pains.
I was doing design, so I can't speak to the actual implementation of the sorting, but the experience had two main components for manual sorting:
1. A user would sort the places she had checked in as 'hate', 'it's okay', or 'love'.
2. She'd also get a list of places (mixed into the same list from 1.) that her friends had checked in, and could answer the same as 1 or answer as 'don't want to try' or 'want to try'.
(Many users would rate 60+ places on their first time using the app, which was a ton of data).
The next thing a user could do is pair up a group of people and find places to go. This was never fully built, but the basics existed: User opens the app and either manually adds friends or the app finds friends nearby, and the app suggests a place within a defined distanced that everyone loves or wants to try. (Or has friends that love that aren't in the group searching, etc)
So it was essentially what you said, except we had our own metrics for love/hate as well as the content from other networks. If a bunch of your friends had checked into a place you'd see it surface quickly, if they identified (in our app) that they hated it it wouldn't surface. There were plans to start doing comparative rating (people who like the three places you like also like place X etc).
CryptoPoops
CryptoPoops
Prime