@aredridel It’s kind of interesting that it took this long for someone to try making an algorithmic feed that mostly shows you stuff you like instead of stuff that makes Engagement Georg comment 431 times.
@vruba I kinda suspect I know how they do it. I think they're actually not personalizing so much as pre-clustering a bunch of stuff, then floating trial balloons to see if you like that cluster. I kinda get the sense that the groupings are more globally coherent than "if you liked X, you might like Y”
@aredridel I think that’s right. They put out a paper about it a few years ago but I assume it’s obsolete now if it was ever the whole truth.
@aredridel Yeah, it seems like a very wise balance. Here’s some discussion of the paper I was thinking of (link at the very end): https://www.gantry.io/blog/monolith-the-recommendation-system-behind-tiktok
@vruba Thanks!