Farcaster Union

The biggest problem with the Warpcast algorithm is that is “spam” means ANY reply with ZERO engagement (no like or reply back). This is very nuanced and difficult to navigate. It is too easy to be labeled as spammy and most users will need to change their behavior when you engage with people who don’t follow you. When you follow each other, the spam rules no longer apply because you are friends.

The biggest problem with the Warpcast algorithm is that is “spam” means ANY reply with ZERO engagement (no like or reply back). This is very nuanced and difficult to navigate. It is too easy to be labeled as spammy and most users will need to change their behavior when you engage with people who don’t follow you. When you follow each other, the spam rules no longer apply because you are friends.

https://p1yefcr.warpcast.tools/api/frames/cold-goose-28

Pi-chi.eth Artist /animeoutcasts
Hosting: /sakura /kyoto /anime-manga /farcasterunion

https://portrait.so/pichi

Wait I thought you only get the label if you're replying to people who don't follow you AND don't engage with your replies AND this is happening more often than the background rate?

Dad, dev • @dune.eth wizard • I'm just zis guy, you know? • Opinions are always my own and don't represent my employer's view on anything.

(I believe Dan said they use 30 signals, some of which are going to be statistical like the above, and others will be machine learning based)

I'm speculating based on how I would have done it... I would take the e.g. 99th percentile of replies data where (not a follower and no reaction/reply).

This would give you a number e.g. 99% of people get at least a 20% engagement rate on their non-follower replies, and if the user is getting a worse rate than that then they're acting against social norms here and are probably spamming so they get a warning.

You can get more fancy and do this calc for each type of engagement and/or the combination, but the idea is that it's an objective measure of what's "normal".