3 ms·

Figuring out suggestions based on preferences and given choices with so constrained options is an easy problem. The "expert" is there to sell you a story. Not

Figuring out suggestions based on preferences and given choices with so constrained options is an easy problem. The "expert" is there to sell you a story. Not that there's anything wrong with that - the story is a part of the experience that many people like - but let's not pretend it's a technical problem that's difficult to solve. I know a bunch of people who work in that buisines (my brother is a chef, gradeschool friend as well, ex. is a restaurant waiter, met a bunch of their co-workers, all worked in Michelin star restaurants etc.) what actually goes on vs what the customer hears ...

majormajor 9y ago
> Figuring out suggestions based on preferences and given choices with so constrained options is an easy problem. Then why do Spotify and Netflix et al persist in identifying artists and songs I've heard of (because of obvious similarities to stuff I like) but have actively ignored (because of any of a number of reasons that don't boil down well into an algorithm), over and over and over again? I still don't want to watch The Office, Netflix. I won't in another six months, either. But I could tell a person why not, and they could factor that into their recommendations. It's like the difference between trusting Waze and trusting a local that says "even though the app says go another half mile then turn right, turning right here is way easier, since if you wait you'll have to then quickly make your way across three lanes of traffic in a hundred yards to make your left, and making the right here gives you more time to get over."
- rubber_duck 9y ago
  Different scale/complexity. You could hand-code waiters decision tree for recommendations and the possible choices end in ~hundreds. Personalized music suggestion has to pick from millions of choices and there's no straightforward algorithm behind it.

majormajor 9y ago

> Figuring out suggestions based on preferences and given choices with so constrained options is an easy problem. Then why do Spotify and Netflix et al persist in identifying artists and songs I've heard of (because of obvious similarities to stuff I like) but have actively ignored (because of any of a number of reasons that don't boil down well into an algorithm), over and over and over again? I still don't want to watch The Office, Netflix. I won't in another six months, either. But I could tell a person why not, and they could factor that into their recommendations. It's like the difference between trusting Waze and trusting a local that says "even though the app says go another half mile then turn right, turning right here is way easier, since if you wait you'll have to then quickly make your way across three lanes of traffic in a hundred yards to make your left, and making the right here gives you more time to get over."

rubber_duck 9y ago
Different scale/complexity. You could hand-code waiters decision tree for recommendations and the possible choices end in ~hundreds. Personalized music suggestion has to pick from millions of choices and there's no straightforward algorithm behind it.