There’s really nothing preventing us from having algorithms on #Nostr like what’s found on Threads, X, Reddit, and so on.
It’s completely doable from a development standpoint, but it's not free to run.
Me personally, if I were to develop a recommendation algorithm for Nostr, I’d probably start with a service that scrapes public relays and stores all notes. Storage is cheap, especially for text, so this is not a problem. The great thing about Nostr is that everything is public, so scraping data is incredibly easy.
Next, I would map user behavior by fetching all user interactions for a specific npub (the npub who wishes to use an algorithm). These interactions could be likes, replies, reposts, follows, and so on. This is to figure out what kind of content the user engages with most.
I would then assign a ranking to each note (could be from 1-10) based on how closely the note matches the user’s observed interests and relationships.
Notes from npubs they follow or interact with often would score higher. The same with notes that resemble content they’ve previously liked or replied to.
Over time, the algorithm would adapt dynamically, so it keeps learning from new interactions. LLMs have certainly made ranking easier.
**The problem:** This is the part that gets expensive (the ranking of notes), and especially when tailored to each user.
nostr:nevent1qvzqqqqqqypzqx84ftc7zrzmk734g69s7c4jjhgjx3us8j0e2uudqewgf0h3gqh0qyt8wumn8ghj7un9d3shjtnwdaehgu3wvfskueqpp4mhxue69uhkummn9ekx7mqpz3mhxue69uhhyetvv9ujuerpd46hxtnfduqzprnywvcy20zl380emtc0llueawuclnjsasfhp7yxksna2jrzr04zxn8uvg