Last Author

Tamás Vicsek began his career as a statistical physicist, studying the collective behaviour of atoms. Later, Vicsek, now a professor at the Eötvös Loránd University in Budapest, Hungary, became intrigued by living systems such as colonies of bacteria. Analogous patterns, he notes, are present in the collective behaviour of many different kinds of systems, as long as the units within a given system are basically similar. Vicsek has extended this idea to people, modelling group panic and crowd dynamics. A few years ago, one of Vicsek's former graduate students, László Barabási, suggested they look at networks. On page 664, the pair, together with Gergely Palla, analyses a network that represents collaborations among 30,000 scientist authors over 12 years, and another network that represents mobile-phone calls among 4 million users over 2 years.

What spurred you to use phone records to study networks?

Mobile-phone calls reflected the level of individual communication in an unbiased way compared with information gained from questionnaires. The extremely large scale of the data collection opens a road to deeper understanding of everyday social processes. The data are completely anonymous and coded — and even the coded data can be provided only to researchers who sign an agreement to use it appropriately.

Why use collaborations between scientists to study networks?

Scientists document their co-authorships quite well and the data are easy to collect. Also, we figured that readers of Nature would be interested in these types of networks.

How did the networks evolve over time?

We looked at a 2-year time period for phone calls, and a 12-year period for scientist co-authors. Large groups within both types of network persisted longer if a portion of the membership constantly changed. But small groups were more stable when membership remained unchanged.

How do you hope to follow up on this work?

Our best hope is that other researchers will use our method and apply it to other data sets. We'd be happy to provide the software free of charge. Also, at this point, the nodes are studied only by their connection — the nodes themselves don't have any specific descriptive features. We'd like to complement the study with more information. For example, we could look at whether the social network groups among scientists in certain fields last longer than those in other fields.