after much research the correct answer to this is personalized pagerank
Login to reply
Replies (20)
WoT
many ways to implement wot. correct answer is personalized pagerank + custom nostr logic
who is going to do the ranking? Clients?
So @Vertex
Isn't that what this guy did already 6 month ago here:
Great job @Pip the WoT guy 🤙
@npub1gcxz...nj5z when build-in-search of @npub142gy...xrj0 powered by @Vertex ? 🙏
View quoted note →
It wont be easy, but we will to teach the world the web always was subjective; you trust platforms to come up with a number, on what the platform beliefs to be real users.
People might be temped to think whatever pops up eventually in the Nostr context is some degraded thing from what we have in the legacy system, but it is probably actually an upgrade
I came to the same conclusion about 1y ago, and that's why @Vertex offered this algo from the get go.
Real-time personalized pagerank.
View quoted note →
Algorithms
Most of our services accept an optional sort parameter, that allows control over ranking behavior to optimize for performance, robustness, or perso...
Goodhart’s Law makes a single good metric impossible.
“When a measure becomes a target, it ceases to be a good measure."
The only solution is so many different measures (ie different one for each user) that it’s unfeasible for spammers to target every one.
Goodhart's law - Wikipedia
Yeah just realized they are using it! Didn’t know that. we will likely do something similar inside our embedded relay
yeah they doing it right
I am gonna do it on the client yeah
hence *personalized* pagerank. It would be different per-user based on their interactions
Wouldn't that require downloading too much data to fit in mobile budget of bandwidth and battery life? Otherwise why don't we have client side search engines?
How is this application different from the needs of search?
The more I think of this, the more I'm convinced that there shouldn't be any attempt to provide metrics of anything.
nostrdb is super optimized for ingesting and storing lots of notes. It’s not really a big deal. Things are incrementally updated and cached.
We also have an embedded local note search. We’re making it to be offline first. we try not to rely on or trust any external third party services, especially for things as important as web of trust calculations.
not to mention in many private relay scenarios, its not like some centralized wot service would even have access to the notes needed to calculate web of trust. Many of them might be things like private zaps or interactions on non public relays.
How does the data required grow as the network grows? If Nostr is the size of Bluesky, how much would my phone need to download?
you don’t need to calculate the entire network, only the people you interact with
Yes but the pagerank of the people I interact with is most useful when I consider the opinions of people I don't interact with.
If I am following a feed in a topic that isn't popular in my following their rank will be low.
Still, happy to watch this experiment unfold.