@liluvou
1. How is that even different from other social media platforms? If you start a Twitter or Facebook account it's the same. You only find your way there because you already know people. Or because Twitter or Facebook know things about you, which is creepy.
2. Then people should complain about the lack of direction towards interesting content, instead of people using languages they don't understand with the implication they don't belong here.

@liluvou I mean, I get the idea for a recommendation algorithm, but that algorithm can only make sensible suggestions if it knows who you are, what you like and/or who you know. I don't think Mastodon wants to go down that path, and I hope the majority of its users don't either.
The price for that is that you have to look for interesting content on your own.

@femmeonabike It's an open field of CS research and I believe there has to be a middle ground between "no recommendation" and ""big brother" (considering mastodon already knows what tweets you fav/boost and who you follow). Better interfacing between instances would be a first step (e.g. logging as a user of another instance in order to look for content), and I think we can already browse local timelines from elsewhere using third-party tools.

@liluvou Hm, now you seem to be assuming that users already follow others and have made some toots (I assumed you wanted users to get recommendations at the time they registered.)
That of course is possible, even if it may not be much data to base a recommendation on.
Could even be implemented as a bot, apart from the main Mastodon development, relieving the core devs.

@femmeonabike I am actually kind of interested in doing that, but I fear a third-party bot may require too much information on users (not to mention doing it in a distributed manner which is another bag of snakes), I’ll have to look at the API

@liluvou Yeah, to make sensible recommendations you'd need to analyse all toots or all relations between all users, or at least all which are available from the instance the bot runs on. I suppose that is technically possible, since a human user can do the same, but you'd have to ask whether it isn't hurting Mastodon etiquette.
A recommendation engine for users should be easy enough to implement, but content recommendations seem pretty hard to me.

@liluvou Actually, I think making recommendations based on who you follow would not even require much scanning:
1. get the list of users you follow
2. get the list of users who also follow those users
3. return the followers who follow the most users you also follow
Done.

@femmeonabike Yes, although some light toot analysis + word weighting scheme would help prioritize specific themes and fields of interest

@liluvou Yeah, that is where I think it gets hard. Really hard. Not only obtaining the data, but also grouping them into topics. You can run a clustering algorithm on them (which is not made easier by the fact that we have many languages here), but you'd have to do a lot of manual curating of that algorithm's output.
Or maybe I'm overthinking this and word frequency counting is enough, but still it's orders of magnitudes harder than follower-based recs.

@femmeonabike
1- It is not that different from other platforms; twitter does recommendation based on your approx. geographical position,other users, and visited pages; facebook does the same but refines what you get the more you fill your profile (which is, I agree, creepy)
2- I am not refuting the fact that people can be scum, especially when they don't think about what they say (and the political implications of it)