Gregory's Server
dansup, the most curious question — does this work across instances, or does it only consider local posts? Recommendation systems are "easy" when you have access to everything you could possibly recommend and you can do all kinds of queries against that data. Not so easy when you have no idea most content exists in the first place.
I'll go another, equally uncharted way though. I want to try decentralized search by centralized identifiers. You'd import your friend list from somewhere, like Twitter or VKontakte, and then some ✨magic✨ will happen and you'll see fediverse accounts associated with those centralized identities. It should, in principle, be possible to make a DHT (distributed hash table) out of instances, and in that you'd store somehow preprocessed centralized IDs as keys and fediverse IDs as values. The idea is that you should be able to find a user on any instance from any instance.
@grishka It uses local-only signals for now, though it wouldn't be hard to use federated content.
The key component is ranking signals in a scalable way, so the smallest and largest instances can take advantage.
Local vs Remote doesn't really matter to a new instance when you have no content, and thus discovery becomes more important.
I'm interested to hear more about your idea, I plan to use FediDB and its growing database to bootstrap discovery in Pixelfed.