Prediction of Social Ties Based on Bluetooth Proximity Time Series Data

Personal ties in most social networks are explicitly declared by its participants, like on Facebook and LinkedIn. Nonetheless, the accuracy of self-declared relationships has been contested. Empirical studies show that behavioral data is much more accurate than self-reported data, as it relies on objective evidence of social link formation. Evidence collected from online interactions have been widely used to elicit social linkage, while physical interactions are rarely exploited to uncover underlying social structure. In this paper, we use proximity data taken from the Bluetooth detection of mobile devices and show that from the analysis of physical proximity, social relationships can be accurately inferred considering time and order of encounters. We also show that moments of time in which there was no proximity are as relevant for social network elicitation as moments of physical proximity. The purpose of this research work is to infer social ties based solely on behavioral proximity data; our experiments substantiate the claim that we are able to infer social structure with high accuracy exploiting proximity clues only.