I have no idea what this means but I love it 💜
Not surprisingly, clustering correlates quite heavily with language, e.g. Germans are in the bottom left. The more international they are, the more up/right they may be positioned. https://canostrical.github.io/contact_cloud/
Discussion
👀
In a perfect world, you should find contacts close to your dot or with the same color cooler than those far away.
It might serve as the basis for recommendations. There is still a lot of spam in the dataset though, and many dead accounts.
Is it possible to visually distinguish between npubs I dont follow, and those I follow?
Is it possible to auto-categorize groups, and auto display a legend?
For example green color is Japanese nostr community, purple is homesteaders etc.
Filtering out those you already follow is already on the ToDo list.
The categorization is a nice idea. I might find a way to use my search index for that. E.g. collect the most used words per cluster or sth.. This might all take a while though.
💪
I wonder if profile info contains folks somehow self-identifying by e.g.
1. Language (e.g. Japanese)
2. A hashtag of interest: e.g. #permaculture #permies #artstr etc
I have a bunch of kind 3 events lying around and used two different methods to visualize the resulting contact list graph.
One is used for the coloring (Louvain) - don’t ask me, why or how it works, but it looks cool.
The other is for the distances between contacts (force direction). Basically it makes contacts repell each other but following relations attract each other, so that in the end subgraphs with a lot of internal followings will cluster.