[ProjectSite (includes live demo app)]
Figure: Visualization of central places in France, Spain and Portugal. Each circle represents a mobile phone tower and its color (the brighter the more central) corresponds to the inverse of closeness centrality (average number of hops to any other person) of the most central persons in this tower. A person is always assigned either to its billing address or most used tower. White lines highlight the social network between the 50 most central persons of each country. In the three insets the distribution of the of all persons and the relation to the used color are also shown
Empirical evidence has proved that social networks are searchable, meaning that a short path from one person to any other person can be collectively found, despite that each individual has only local information of the entire network. The structure of empirical social networks that allows these phenomena has not yet been completely uncovered, and this knowledge is important for communication systems and models of information spreading. In this work we evaluate diverse decentralized search strategies in social networks of over 25 million phone users from three countries. While the city of the target person is reached within a few hops almost independently of the origin and the routing strategy, we show, for the first time, that within cities, the strategies based on geographic information completely fail to reach the target. This failure occurs because the relationship between the social networks and their geographical space does completely change within cities: social groups are geographically scattered across the city, as opposed to what happens at country scale where communities nearly preserve geo-political borders. We present structural evidence of why routing strategies based on the network’s community structure are successful within cities.