 I think a reasonable solution to dealing with spam account mitigation is to build out and embrace user level invites Allowing users to invite their friends and leveraging their "trust score" to apply our Autospam detections accordingly. This would solve the UX issue, while still allowing for strict vetting of new users. What do you think? (Boosts & feedback greatly appreciated) 11 comments
@dansup I’m pondering how gameable it might be. How would you calculate the trust score? Imagining a network of bots who all invite each other…  @dansup When I joined fediverse, there were exactly 0 of my friends (from other networks and real life) here. @po3mah I understand, the "trust score" is only used to determine what Autospam mitigation actions to take, it won't prevent you from inviting friends, rather its just a "signal" on how we can trust your invited friends based on how good or bad you've behaved! It's actually pretty simple when you break it down, and we will make this clear in the docs and Help Center! It assumes that people will be reasonable and bad actors don’t game the system. Spam is a surprisingly hard problem to solve when using simple rules. @DevWouter @sam That is true, but we have also considered that, we will perform "shadow detections" that are silent and undetectable until a human admin can verify, and silently restore access That would work. It sounds like you slow down the system. And spam it an often focused on short term gain. Another approach I read somewhere is that new accounts get throttled on what they can do until they are considered “proven”. @DevWouter @sam Yeah that is what discourse does, and that does make a lot of sense. We'd have to do this carefully though, so as to not discourage legit users, so I think allowing them to do actions and reviewing them later is ideal UX wise. Wdyt? Depends on the persona. Some users join and want to be very active from the very first second. Others will join, check things out and have less interaction. (I will ignore migrating accounts) Then there are ways of limiting them. Do you want to prevent interaction or delay interaction? You can also make a distinction based upon the way they interact. Spam can be broadcasted or 1-on-1. @dansup |
Gregory's Server
@dansup I agree. not bulletproof, but effective.