I made a Lemmy instance with a custom algorithm that keeps only the top 20% most unique (=interesting?) posts. It does this by calculating a similarity score between every post on my instance and all posts that came before it. The top 80% of posts with the highest self-similarity get removed instantly.
The idea would be that this allows me to cut through the noise that’s running through the communities, similar to how xkcd-signal attempted to do 20 years ago.
The instance is mostly meant for reading, not posting. So it has a very open federation policy (for now).
If anything, this is experimental. So please let me know what you think! You can see the type of stuff that gets removed in the modlog (https://lemmy.coffee/modlog).
I do very few things explicitly, I just punish self-similarity in a very specific way. I guess posts with actual text in the body are just more unique, given all previous posts on the instance.
Maybe using the filtered posts as a base in combination with some client side keyword blocking will be useful? The keyword blocking would be much more individual for each user.