Clustering temporal networks


Before I started my PhD in Sune Lehmann's Lab, I did a cool little project with him. Following his and Vedran Sekara's PNAS paper on social cores he wanted me to build a data visualization that showed the temporal dynamics of these social cores (or "groups" is you will). Over a semester I did just that, and I was even so lucky to win first place in Science Magazine's Data Stories competetion with this visualization as my entry.

Here is the data visualization.

So when I started my PhD, we figured it would be nice to revisit this and make the community detection part (finding the social cores) a little better. We teamed up with Martin Rosvall, author of Infomap, and developed a good flow-based algorithm for community detection in temporal networks. We put it in Physical Review E (also on the arXiv), and integrated the method into Infomap with the flag --multilayer-js-relax-rate. Here's a figure from the paper, showing some social groups at the Technical University of Denmark expanding and contracting over a day: