this post was submitted on 18 Jul 2023
575 points (98.5% liked)
Fediverse
28544 readers
380 users here now
A community to talk about the Fediverse and all it's related services using ActivityPub (Mastodon, Lemmy, KBin, etc).
If you wanted to get help with moderating your own community then head over to [email protected]!
Rules
- Posts must be on topic.
- Be respectful of others.
- Cite the sources used for graphs and other statistics.
- Follow the general Lemmy.world rules.
Learn more at these websites: Join The Fediverse Wiki, Fediverse.info, Wikipedia Page, The Federation Info (Stats), FediDB (Stats), Sub Rehab (Reddit Migration), Search Lemmy
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
General popularity is not a good metric IMO. If I like a community, then it shouldn't matter if a million people like or it's only me and my cousin. If the community likes the content, I want to see it.
It's trust between the members of a community.
However, weighted sorting is not a solution too, upvotes counts are not linear. Maybe, quantile sort?
What's liked by the general population is a good metric for providing general stuff to the general population and that's what we're talking about in All.
That average can however deviate a lot from the sweet spot for some people, quite possibly a large minority (even the majority depending on how concentrated or not people's tastes are around it).
Something that looks at your previous choices (or even generally stated choices in the form of communities you subscribe to or block) similarly to what some search engines and some social media sites will do, can shift that toward more your own specific tastes, but that's computationally more expensive and requires more users and more user data to get better results (basically it's finding certain kinds of users and local minima which are more satisfactory to them).
I suspect something like an AI solution (not LLM, just a much simpler neural network) running on your own device that tries to predict what you're going to click on and learns with what you do (or not) is the only way for a personalized "no fluff on my feed" solution, but that's for apps running on top of Lemmy, not the Lemmy engine.