Scaling Up

Our previous results used a sample of 60,000 user profiles so that we could compare our performance against that of a range of popular community detection techniques, some of which do not scale well. Here we show the results of our technique applied to a much larger dataset, 1 million profiles. We show that the results continue to be plausible, both using four modalities (that is, with followership symmetrised) and five modalities (with followership directed).