1y ago

Hot take: developing an open source For You algorithm for mastodon that shows you posts based on your likes inbetween every other chronological post would be great

If we had an open source algorithm for Mastodon/Pixelfed that learned based on the words in the post and image/video we could have a Following + For You feed that showed you all the posts from people you follow and you could choose to see, say, 1 recommended For You post after every 3 posts from your Following feed. With the option to disable For You posts completely or tweak how often you see these.

Discovering new people to follow on mastodon/pixelfed isn’t great (hashtags are rarely used and make posts look ugly) so I still occasionally use twitter because I often discover new indie animators/gamedevs showing off their project making it really nice to browse the For You feed.

13 comments

This seems like it could have negative side effects. For example if you like posts about negative stories and the algorithm interpreted you being supportive of the post as you appreciating negativity it could work to foster a falsely over negative world view in you sending you into a worse mental state. Or if you dislike a post you might not agree with it could lead to you only seeing posts that reinforce your beliefs to the point of radicalization.
- That’s not really how most recommendation algorithms work. A good basic algorithm would be showing you content that the people you follow are following or showing popular posts from a certain category that you tend to like

This and the way Mastodon handles likes/favorites are the biggest things holding them back right now, IMO. People don't want to bother with a social media they actually have to put in the work to curate themselves, hell the algorithm is one of the main reasons why TikTok is so successful. Also, by having favorites (which is... weird and confusing? Just call it "likes", you don't need to reinvent the wheel here). Not only do people get confused between favoriting something and bookmarking something, but since it's only counting the favorites from your instance, it makes it look like posts are getting barely any interaction, giving this "screaming into the void" feeling that will cause a lot of users to lose interest. The number of ~~likes~~ favorites needs to be cumulative across instances.
- You’re 100% right. The people who are already on these platforms are the ones that don’t like any algorithm at all, so they tend to be combative about its implementation. I don’t think there will be mainstream appeal until it is easy and fun to use for the average person

100%. Simply keep chronological order for the boomers on there and give us a For You algo for actual interesting content. I don't want to keep getting gaslit that it's supposedly better and more natural that way--It's boring as hell.

There are effective ways to be discovered on Mastodon: by being boosted by others and by using hashtags.
If a user wants to be discovered they should be using hashtags. If they're not prioritizing discovery then they shouldn't. And the "consumer" (for lack of a better word) can follow those hashtags, so they appear automatically on their home timeline.
If a user wants to make their followers to know about a person then they should boost their content. That's how relationship works on real life: Your friend sees something cool and snows it to you.
- I know how reposts work. I know how hashtags work. They’re not great way for discovery especially for discovering smaller accounts. I constantly get recommended tiny accounts posting their gamedev or indie anime work through the algorithm. We can have chronological and algorithm on one feed so its the best of both worlds
- word clouds will find when a corpus of hashtags is similar in meaning. If you use only hashtags, that's like experiencing a grocery store or farmers market via an ambassador who cannot see the serendipitous shops that are nearby, the things frequently seen together. It's like shopping in an app and never visiting the grocery store itself. Having a precision following list means you can't experience going to a library and browsing shelves until something catches your eye - serendipitous search is fundimentally different from subscribed/reposted delivery, or even keyword search.
  computers and digital space don't natively have the metric for which hashtags are closer, so they have to crunch the numbers to help figure out which books are closer to other books. Otherwise it's like entirely separate universes that you'll never ever find, like if you never knew a word that would lead you to a community of much more words and concepts and free thought.
  
  I guess the thing I'm trying to describe is browsing. https://www.youtube.com/shorts/5GggiXMaqDE

Is this a showerthought or a pitch meeting?

Stuff like this could be added to misskey, Firefish, or whatever other fediverse platforms there are. No need to implement it on Mastodon & piss off the entrenched community that wants no algorithm.
As for the algorithm, im imagining a few dedicated relay servers to sift through mastodon server data using a dedicated llm (thinking like llama or hugging face? I don't know much about the technicals) to sift through and rate posts by engagement, word choice (evil, demonic, eugenics, crypto, etc) and rate each post topic collectively based on how virtiolic, how much engagement, how many posts, how similar to other posts (for spam), & the topic category & subcategory based off the respective language.
Users who activate the feature would have likes, saves, & etc stored on the client & a hash sent or something with the potential profile focus of the person.
The server weighing the details will prefer to return less seen & posted content very specifically tailored & occasionally more notable & engaged posts that are more general & less charged.
I think the general idea is interesting, & I would try to build it myself (for those who say stop whining & make it yourself), but im here grinding through kaggle learning & udemy so it'd be a while lol

ehhh idk, just use hashtags or look on specific instances if you wanna find new stuff

13 comments