Location Sharing Behavior: Analysis and Models

A range of mobile applications allow individuals to create geo-located content using location-based services like Foursquare, Facebook and Twitter. This location-based sharing creates new opportunities for users to associate their life events to geographic locations, e.g. places they have been feature in Facebook and share their locations in a social context with their connections on the social network to inform them about their whereabouts. But these opportunities present a clear risk to user privacy and location privacy. And yet, many users continue to voluntarily share their location information. In this thesis, we aim to study the factors impacting location sharing behavior toward providing a foundation for future adaptive location privacy systems which can help users decide whether it is safe to share their location or not. Concretely, we study a unique Twitter-based dataset of (i) users who always share their location, (ii) users who never share their location, and (iii) users who selectively share their location. We conduct a data-driven analysis of location sharing via multiple factors including the time of the Tweet, the content of Tweet, and user profile features. Based on this data-driven analysis, we investigate whether we can predict whether a Tweet will be tagged by a user with geo-location or not, a key step for enabling an adaptive location privacy system. We create a global classifier for all users to uncover the common driving factors for location sharing. We also build per-user individual classifiers to improve the prediction performance and to view the users in a spectrum of predictability. We achieve an accuracy of 70% for the global classifier and an accuracy greater than 90% for more than 60% of users. We observe that features like the users social status, the source of the Tweet, whether the Tweet has a mention or not, and the textual content of the Tweet are the most important features. These observations imply that users are conscious of their online visibility and social connections while geo-locating and also that the usage of mobile devices promotes location sharing. We also conclude that most users are highly predictable in terms of their location sharing behavior and thus our work creates a substantial groundwork for future data-driven location privacy systems