On Mastodon, if you have an account on instance X, you can follow someone who is on instance Y. It creates a connection: X -> Y. If there are a lot of such follows, weight of this edge will increase, attractive force between points will be higher.

Original explanation on the page of Kaggle dataset:

"active users" graphs: For each instance, we consider the set of the 10K most recently active users. Then, for each user of an instance X, we consider the list of the users they follow, and add 1 to the edge from X to Y where Y is the instance the followed users. The weight of the edge from X to Y thus encodes how much the content seen on instance X is generated in instance Y. Note that this graph thus contains self loops.

I've tried to layout this dataset in Gephi, but it was a classic hairy ball - everyone is connected to everyone, amount of edges is too high comparing to number of nodes. Then, I've filtered out all EN instances and suddenly got a meaningful picture:

What can we see? If English-speaking instances are ignored, German, French and Japanese languages are most common across Mastodon. Japan and Korea don't hang around much with other folks, while French, German and Spanish instances are quite interconnected between each other.

Size of nodes depends on centrality, post about centrality of Peertube instances is here.

Same, but Fruchterman-Reingold algorithm instead of ForceAtlas 2:

Mastodon active users dataset can be downloaded here: https://www.kaggle.com/datasets/marcdamie/fediverse-graph-dataset-reduced