On the predictability of recurring links in networks of face-to-face proximity

This paper focuses on the predictability of recurring links: These links are generated repeatedly in a network for different forms of social ties, e.g. by face-to-face interactions in offline social networks. In particular, we analyse the predictability of recurring links in networks of face-to-face proximity using several path-based measures, and compare these to network-proximity measures based on the nodes' neighbourhood. Furthermore, we show that the current tie strength is a good predictor for this link prediction task. In addition we show that the removal of weak ties improves the predictability for most of the considered network proximity measures. For our analysis we utilize three real-world datasets collected at different scientific conferences using the Conferator (http://www.conferator.org) system.