[spamizbad]

Having read the paper, what's unique about Bytedance's approach is how relatively simple it is at its core - obviously there's a lot of complexity around it to do it at scale, but I feel like it's simpler than the social-graph based approaches.

[mrfox321]

It's simpler intuition but more complex from a data / ml perspective.

Their algorithm is really built around their features. Specifically, temporal representations of user interest:

https://ieeexplore.ieee.org/document/9458799/

The features used by their algorithm tells you what a user is interested, historically.

Contrast this to Meta, which uses the social graph as their features. Imagine features like the number of times a user likes another author's / cluster's content.

Tiktok will serve you $TOPIC because you have $INTERACTED with $TOPIC historically.

Meta will serve you $TOPIC because you have $INTERACTED with $PEOPLE who post $TOPIC, historically.

Meta only coincidentally gives you what you like.

Tiktok knows what you like.

This is the difference. This is why IG is losing.

[xnx]

That's a crazy design choice by meta if true. The interests of those in my social graph have very little connection to my interests.

[mrfox321]

It's because they originally built their recommendation system to recommend friends and their content. Here, the social graph makes complete sense as the foundation for their simple search algorithm. But as they expanded their recommendation capabilities, the features stuck around. It's the same reason why tech debt accumulates. Data sticks around in the same way code does. But data is even higher friction, since it's a superset of the code.