Nature 489, 537–540 (2012)


Popularity may be an attractive trait, but when it comes to evolving networks, Fragkiskos Papadopoulos and colleagues have determined that it competes with familiarity. The emergence of scaling in networks is often cast within the framework of preferential attachment — the idea that popular nodes accrue links more frequently than their poorly connected cousins. This assumption recovers the power-law scaling of connection distributions in several real-world networks, but falls short of accurately describing their evolution.

Papadopoulos et al. have developed a model that plays popularity off against similarity — demonstrating that evolving networks optimize the formation of new attachments by trading links to popular nodes with those that connect similar nodes. The most intuitive example is the Internet: a model network that preferentially creates new links to Google and Facebook, based on their popularity, will omit more subtle links, such as those connecting the sites of cycling enthusiasts to their favourite Tour de France forum. But the framework is also applicable to the less obvious evolution of social and metabolic networks. Perhaps most importantly, it holds promise for accurate prediction of new links in evolving networks.