Parameter Optimization for Deriving Bluetooth-Based Social Network Graphs.

Pervasive technologies such as Bluetooth (BT) are capable of detecting close proximity. As a result, they are increasingly used for deriving social networks. However, the validity and reliability of the inferred networks is questionable as evaluation procedures are often omitted. In this paper, we consider the process of deriving and evaluating a Bluetooth derived network as a parameter optimization problem. Using the BNEA algorithm, we investigate the effect of the number of detected connections, time window in which these are detected, and the direction of the resulting connection. Our results confirm the importance of conducting a throughout evaluation procedure when deriving social networks based on BT data. Going through the parameter optimization process, we are able to increase the accuracy of the derived BT networks by a maximum of 10%, compared to deriving the networks without it. Our outcomes indicate that reducing the false positives can be achieved by setting a particular connection_weight. Furthermore, with the window_size parameter we show that more BT observations does not necessarily mean more accurate networks. With respect to the connection_type, we observe the accuracy of deriving undirected networks is higher than the accuracy of deriving directed networks. Finally, based on the outcomes we are able to come up with a set of recommendations for the future developers of similar BT data collection systems.