Probabilistic range monitoring of streaming uncertain positions in geosocial networks

We consider a social networking service where numerous subscribers consent to disclose their current geographic location to a central server, but with a varying degree of uncertainty in order to protect their privacy. We aim to effectively provide instant response to multiple user requests, each focusing at continuously monitoring possible presence of their friends or followers in a time-varying region of interest. Every continuous range query must also specify a cutoff threshold for filtering out results with small appearance likelihood; for instance, a user may wish to identify her friends currently located somewhere in the city center with a probability no less than 75%. Assuming a continuous uncertainty model for streaming positional updates, we develop novel pruning heuristics based on spatial and probabilistic properties of the data so as to avoid examination of non-qualifying candidates. Approximate answers are reported with confidence margins, as a means of providing quality guarantees and suppressing useless messages. We complement our analysis with a comprehensive experimental study, which indicates that the proposed technique offers almost real-time notification with tolerable error for diverse query workloads under fluctuating uncertainty conditions.