by Keith Coleman & Jay Baxter
Instead of asking 'is this note popular?', the algorithm asks 'is this note helpful to people who usually disagree?'. It utilizes matrix factorization to identify user clusters and only surfaces content that bridges the divide between polarized groups.
Core Principles
- 1.Identifies 'polarized' clusters based on past rating behavior.
- 2.Requires a 'Bridging Signal': Positive ratings must come from diverse viewpoints, not just one side.
- 3.Prioritizes context over censorship: Add context to misleading posts rather than deleting them.
- +1 more...
"We actually look for agreement from people who have disagreed in the past... that's what makes the notes so neutral and accurate."