[marginalium]
Algorithmic Ranking Is Unfairly Maligned
27 Jan 2025
Algorithmic Ranking Is Unfairly Maligned. Good bit on why Netflix is so bloody useless:
In the long-long ago, Netflix had star ratings … Nowadays, you get a disorienting set of categories like DARK COMEDIES ABOUT ITALIAN FEUDALISM and LIFE IS SHORT—WATCH IT AGAIN and THINGS YOU’RE IN THE MIDDLE OF, HELPFULLY PLACED IN A INCONSISTENT LOCATION. Instead of star ratings, there are “match percentages”, but you have to interact to see them and they always seem to be 98%.
Then:
Netflix realized a bunch of things:
- That they needed to concentrate everything on increasing subscriber revenue. And that the main goal of recommendations should be subscriber retention, or making sure people don’t cancel.
- That the things people rate highly aren’t always the same as what they actually watch. It’s cool that you gave The Seventh Seal five stars. But after a long day at work and finally getting the kids to bed, are you really going to choose Andrei Rublev over The Great British Bachelorette and the Furious 7?
- That to retain people, you need to get them started watching new stuff. Lots of people want to watch Friends, so Netflix will pay $100 million/year for Friends. But if you just join, binge every episode of Friends, and then cancel, that’s bad. However, if the Friends button were to—say—randomly shift around in the interface, maybe while hunting for it you’ll get hooked on some other (hopefully cheaper) shows and stick around longer.
- That beyond your explicit ratings, there are lots of implicit signals like what you watch, what you click on, what devices you use, and how long you stop scrolling when shown different kinds of thumbnails. These implicit signals are more useful than explicit rankings when predicting what to show you to keep you subscribed.
- That many people don’t want to rate stuff. And (I speculate) that this provides a convenient excuse to drop the whole star rating system and replace it with the “whatever the hell order we want” system that prevails today, where the match % means nothing and promises nothing.
Anyway. Makes some interesting points toward the end, about how if capitalist ranking is broken, addictive against your interests, not for them, then either we do no algorithms and rely on other curation tools like RSS etc, or we bake user control in to the algorithms. Not really that groundbreaking, but it a good prompt to think about what you’ll do about algorithms since they’re probably just going to be more prevalent as we try to inject AI into everything.
Anthologies: Gratification, Digital Architecture, On (Un)happiness, On Being Fruitful, Noetik