The original data had 21,000+ features. I used an algorithm to reduce the dimensions to 2 but keep a similar structure (so similar communities are close dissimilar communities are far away).
So the axes don't really mean anything in particular.
Probably a webgl problem. I had to use ungoogled chromium to open the page. I think it works on regular firefox too.
Yeah that sounds like a good idea so you can see how connected local communities are. Probably makes more sense to use original dimensions so no extra information is lost.
Something that I find interesting is how close the central clusters of beehaw.org, slrpnk.net, and lemmy.blahaj.zone are together. If you only highlight those instances then you see how close their communities tend to be.
Total communities: 2986
Total users: 21934
So the dimensions were reduced from (2986, 21934) to (2986, 2)
Edit: Also yeah it is using Umap for the algorithm and it does do something pretty similar to what you described.
I was somehow able to get both a picture and url added and it looks much better. Thx.
If I can figure that out I would definitely do that.
Either the people in !steamdeck@lemmy.world are pretty horny or its an artifact of the dimensionality reduction and means nothing.
Edit: Actually it could also be that it just didn't collect enough data on that community and the most recent person was also active in nsfw communities. I was only able to get back 14ish days in the data for lemmy.world. They produce way to many comments and I got kicked out early.
Yeah pretty much. I wanted to see communities that had similar people that commented because I thought that would be a good way to see if there were similar kinds of discussions were happening in those communities.
For example most of the red dots to the top right are nsfw communities and it was able to clump like that because the people that comment in those communities tend to comment in the other nsfw communities as well.
edit: left -> right
I didn't measure activity for this map. Each dot represents a community. I only used the communities that were on the top 35 instances (except lemmings.world which it couldn't grab any comments for.)
Yeah I've noticed there aren't many clusters that encode specific ideas (there are a few like the anime, nsfw, or sometimes instance level clusters). Most of it just seems to be a blend. Sorta disappointing.
~Anti~ ~Commercial-AI~ ~license~ ~(CC~ ~BY-NC-SA~ ~4.0)~