I mean 1) 'stable matching' is not how anyone would describe the experience of using IRL meetup apps
2) at best gale-shapley is being used for ranking, not for preference inference; IMO they removed the 'have you met' feature (guessing because users found it invasive and hated it), but would be interesting to use it for scheduling meeting slots, a resource problem more similar to med school matching. The idea of using math to meet a mate goes back at least to kepler https://www.npr.org/sections/krulwich/2014/05/15/312537965/h...
3) gale-shapley is old science about how to rank given preferences, but the actual interesting question is how they're detecting preferences. What are the factors? Are some factors excluded? What can hinge (and their cousins at match.com and elsewhere) detect about a person from their profile + interactions? Are they using image analysis / NLP on profiles / chats?
(they're definitely not just using the user settings toggles for preferences; at minimum, they must have a global rank for showing popular profiles at the top of the stack. Also it makes sense this stuff is secret, it's a liability landmine)
I was wondering the same thing re: #3. Some of it is answered in the article but it reads like it was written by a pop-tech blogger and obviously very light on details. I'm not sure where the linked article gets its information, but it seems to suggest that Hinge might be doing some kind of image analysis, maybe:
> Your likes also play into it. The more you like certain types of people, the more Hinge learns about your preferences, as founder Justin McLeod told British Vogue: “It gets better and more accurate the more that you reveal your tastes.” It’s kind of like if you look at the past ten people you’ve dated or fancied, you’ll eventually start to see a pattern emerging, so you’ll have more success (in theory) if you start to seek these types of people out. I.e what’s the point in me going for short blondes when I know I’m only really attracted to tall brunettes? The answer is there is no point, and Hinge knows that.
And why wouldn't they be? It seems like an incredibly easy target for aligning visual preferences without having to ask.
2) at best gale-shapley is being used for ranking, not for preference inference; IMO they removed the 'have you met' feature (guessing because users found it invasive and hated it), but would be interesting to use it for scheduling meeting slots, a resource problem more similar to med school matching. The idea of using math to meet a mate goes back at least to kepler https://www.npr.org/sections/krulwich/2014/05/15/312537965/h...
3) gale-shapley is old science about how to rank given preferences, but the actual interesting question is how they're detecting preferences. What are the factors? Are some factors excluded? What can hinge (and their cousins at match.com and elsewhere) detect about a person from their profile + interactions? Are they using image analysis / NLP on profiles / chats?
(they're definitely not just using the user settings toggles for preferences; at minimum, they must have a global rank for showing popular profiles at the top of the stack. Also it makes sense this stuff is secret, it's a liability landmine)