This is a community built open-source clustering visualization for subreddits across the platform.
It can be used by people to
find related interests
see how reddit communities connect to each other
look for popular communities to explore new interests
Note from the author:
The creator of this visualization collected over 176M unique comments left by users on the website and computed Jaccard Similarity between subreddits.
Then treated relationships between subreddits as a graph clustering problem.
Once clusters are computed, treated them as a graph layout problem and created an SVG file.
Once the SVG file was created it was treated as a dual problem of WebGL rendering and streaming SVG parsing, and the author created his own WebGL renderer and streaming SVG parser.
This was created for an average internet user who would like to find similar interests or look for new popular intererts.
Choose any subreddit which you like, say funny.
The graph will show all the related subreddits which you can explore.
We can see examples such as gaming, pics and politics here
Select any subreddit you like
Click the Show related button on the screen
3. The graph looks throughout the whole network and finds all the related subreddits to the one we selected
The graph has got a large subreddit section you can select any community from there
And from there you can select the street view from the graph
After which the graph takes you to a large section where you can explore all the subreddit in an immersive fashion
As you zoom in it shows more items that are invisible in the outermost view.
The names vanish when you zoom in
The subreddit pops up when you're looking at it.
There is also a smaller second island at the bottom of the chart with no labels. WARNING: It's of all the NSFW subreddits. The creator did not do this on purpose create a separate island for this content but users have alternate accounts for sharing and posting the posts. The clustering algorithm automatically created 2 different islands/clusters.
Great insight into the structure of the website/data
The regions sharing boundaries sometimes have no connections at all.
It can be said that Weed Drugs and Crypto are more of a giant cluster
And Sports Fitness Soccer can be another cluster.
Since Distance parameters have been used to categorize them we can’t do hierarchical clustering
Could add a list of all the connecting communities when you click on one it makes it easier than zooming to see all of the names. The red space could be a list.