No matter which sort you use (except for new), content is recommended to you by activity. Depending on the sort (active, hot, top) it uses a slightly different mixture of votes/comments/time since post to determine the order.
The only exception is scaled, which boosts a little bit midsized communities, but still doesn’t manage to improve visibility of niche ones.
If lemmy is to truly start having active hobbyist communities instead of being 95% lefty US politics, Shitposts, and some tech stuff, it needs a sort that takes into account the user’s engagement.
For example, if I upvote / comment often in a community, there should be an option to have posts from the community be boosted in my feed, even if it’s a tiny community.
Let’s say I’m subscribed to !world@lemmy.world and !news@lemmy.world because I want to occasionally see news. However, I’m also subscribed to a couple hundred other communities, some of them who don’t manage to get more than a couple upvotes on their biggest posts. And whenever I see them I’m replying/upvoting because I’m passionate about that topic.
My feed shouldn’t be 95% c/news and c/world because those are the most upvoted and commented. I shouldn’t have to scroll down hundreds of posts to find “big” posts in small communities I interact with at any opportunity I get.
That’s why I think it would be beneficial to lemmy if the sort/algorithm took into account your engagement in a way.
It doesn’t have to be complicated, you can have a single number “engagement score” for every community calculated with a basic formula, and that number is used as a boost to the community.
I’m aware that there are some examples of successful niche communities on lemmy. But that’s mainly because either a significant chunk of the lemmy userbase is into that niche (let’s face it the lemmy community is not a representative sample of the world population, we tend to be very similar people), or because the posts on it are simplified image/video type posts which appeal to people who don’t know much about the subject.
Good algorithm should:
https://quiblr.com/
This looks cool, but I would prefer to not pass my credentials to a third party, especially for features that should probably be in the default client.
I think to trying meet all of these without compromises (such as privacy and performance) is basically impossible. How would one boost engagement on positive emotions or personalize without large data mining efforts, model building, and running text classification on every comment or post?
I agree they are good aspirations.