The prevalence of social network sites and smartphones has led to many people sharing their locations with others. Privacy concerns are seldom addressed by these services; the default privacy settings may be either too restrictive or too lax, resulting in under-exposure or over-exposure of location information.
One mechanism for alleviating over-sharing is through personalised privacy settings that automatically change according to users’ predicted preferences. This talk will describe how we use data collected from a location-sharing user study (N=80) to investigate whether users’ willingness to share their locations can be predicted. We find that while default settings match actual users’ preferences only 68% of the time, machine-learning classifiers can predict up to 85% of users’ preferences. Using these predictions instead of default settings would reduce the over-exposed location information by 40%.
This work has mainly been performed by my PhD student Greg Bigwood, but Tristan will be presenting the paper (at the AwareCast Pervasive workshop) because Greg will be busy in St Andrews graduating!