Friend networks graphing

[towerless]

I've been recently thinking about various graphical ways to represent online communities, those in LJ in particular. There is a bunch of tools that build one sort of graph or another, but none seem to provide a good view of a community the user in question is a part of. While investigating this problem I've built a prototype tool that tries to

Comments

[avatar]

My thoughts are that you should advertise and release this tool here so we can use it.

[st_arbirix]

int * thoughts(void)
{
        return( !(I don't currently plan to distribute) );
}

[st_arbirix]

Since it's too lofty to assume that you're the center of this cloud I'd think what you'd want to do is travel two steps and recenter the cloud based upon the person with the highest number mutual friends among all the people within their reach. It complicates the excercize by an order of magnitude but gives you a better approximation for any 2-deep community you may be in. As far as communities go I know that I am certainly not the center of the speghetti pile of friends that I have within a two separation limit.

[st_arbirix]

Ack, forgot something...

You'd also want to make sure that you're counting yourself not only as part of a community centered around these new clusters but that it's the community you're primarily involved in.

I have one friend who is for the most part completely unrelated to the rest of my community. Among my other friends the people fall into two categories where one is twice the size of the other. Oddly enough, the smaller community has ties to the larger that are not just through me thought the friends happen to be geographically and historically distinct.

The whole exercise would be much more accurately (but more complicated) if we could weight the associations between people. For instance: by how often they post on each other's journals.

[towerless]

good points, thanks!

[hythloday]

I'm not sure a satisfying representation can be had with such a low amount of data (two bits per relationship).

If XFN had a more serious following, or lj support, then you could, for example, weight the springs, and then solve the graph as a constraint problem, which would clump the users with the closest relationship. As it is, this isn't possible, and your representation suffers for it: "maruse4ka" (top right) looks far more like an outsider than "lenoyka" (middle left), despite both of them having 2 links. I suppose you could organise your cloud into concentric circles, with the number of links the node has determining what circle they live in: have you investigated that?

[st_arbirix]

Wouldn't weights based on cross-posting help? I'm thinking that that's something that can be automated easily enough. It'd be much simpler than asking people to weight their relationships.

[hythloday]

Any kind of weighting would be better than none, yes. I don't believe that you can automatically get the amount of cross-posting without violating the TOS, though.

[st_arbirix]

Ew. Heh. I should probably read that...

[taitdabogan]

this all reminds me of Planarity