Inferring geographic coincidence in ephemeral social networks

Abstract

We study users' behavioral patterns in ephemeral social networks, which are temporarily built based on events such as conferences. From the data distribution and social theory perspectives, we found several interesting patterns. For example, the duration of two random persons staying at the same place and at the same time obeys a two-stage power-law distribution. We develop a framework to infer the likelihood of two users to meet together, and we apply the framework to two mobile social networks: UbiComp and Reality. The former is formed by researchers attending UbiComp 2011 and the latter is a network of students published by MIT. On both networks, we validate the proposed predictive framework, which significantly improve the accuracy for predicting geographic coincidence by comparing with two baseline methods. © 2012 Springer-Verlag.