i think the fundamental issue with a feature like themed lists/groups for suggesting followers is that what you like or do has, in my experience, very little to do with what kind of person you are or if I'll like hanging out with you. I like game development & indie music, but guess what, there's a significant contingent of people who also like those things who are just fucking terrible.

the best indicator of whether or not i'll like someone is if my friends are friends with that person. the problem with this approach, though, is that sites structured like twitter and mastodon don't track *friends*, they track *followers*, and interpreting following as friendship leads to this awful one-sided clumping behavior where the people that get recommended are people that have a lot of followers, so they get more followers, so they're recommended more, etc

I think maybe the solution to this problem is to base recommendations on how many of your friends follow *and are followed back* by someone. there's probably also systemic issues with that, though. it might just lead to like, insular cliques for instance, though i think that's better than the clumping of social capital you get with the old model

@lycaon
What about taking the number of symmetric follows, then weighing that by the number and average length of reply volleys?

(Alternately, compare language use in public posts, ex. with word2vec + TFIDF.)

@enkiv2 that sounds needlessly complicated and also bad