From Signal to Social: Steps Towards Pervasive Social Context

The widespread adoption of smartphones with advanced sensing, computing and data transfer capabilities has made scientific studies of human social behavior possible at a previously unprecedented scale. It has also allowed context-awareness to become a natural feature in many applications using features such as activity recognition and location information.

However, one of the most important aspects of context remains largely untapped at scale, i.e. social interactions and social context. Social interaction sensing has been explored using smartphones and specialized hardware for research purposes within computational social science and ubiquitous computing, but several obstacles remain to make it usable in practice by applications at industrial scale.

In this thesis, I explore methods of physical proximity sensing and extraction of social context information from user-generated data for the purpose of context-aware applications. Furthermore, I explore the application space made possible through these methods, especially in the class of use cases that are characterized by embodied social agency, through field studies and a case study.

A major concern when collecting context information is the impact on user privacy. I have performed a user study in which I have surveyed the user attitudes towards the privacy implications of proximity sensing. Finally, I present results from quantitatively estimating the sensitivity of a simple type of context information, i.e. application usage, in terms of risk of user re-identification.