Some more rough notes on ...

This viz is one in a series I've been doing - you can see some more, including Twitter social graph, Dopplr city travel clusters and Delicious tags here:

I'm working on this series for an eventual blog post about how turning data into graphs makes it possible to do all sorts of clustering, visualisation and analysis. Because of all the work that's been done in the past on analysing the web-graph (e.g. PageRank) and the social graphs (e.g. 100s of academic papers written on Twitter and Facebook), if you can turn your data into a graph then you can borrow those methods for other domains.

For example, the travel-graph on Dopplr is made by taking the start-city and end-city of each trip taken by a Dopplr traveller. In this graph, cities are nodes and trips are edges. Then we can calculate which cities are the most central to the Dopplr network, ignoring geography and focusing only on social activity:

I call these graphs "social-object graphs". Like social graphs, they show relationships made by people's activity or opinions, but the nodes are social-objects instead of people. c.f. by +Jyri Engeström

While writing this post, the SXSW panel picker was launched for 2012 and I noticed that there are lots of common tags between posts. These can be used to make a co-occurrence graph. My tool of choice once I have such a graph is Gephi, the "photoshop of graphs". It can perform statistical analysis, filtering and algorithmic layout, and that's what I did to produce this SXSW viz. The nodes are sized by how "important" they are to the network. Commonly co-occurring nodes cluster close to each other, leading to a nice readable layout - if two nodes are far away, they are unrelated (e.g. UX and developer talks opposed to social media and marketing talks).
Shared publiclyView activity