Using Sparse Digital Traces to Fill in Individual Level Mobility Timelines
Predicting individual-level mobility patterns is an imperative part of ubiquitous computing, in growing real-world applications like transport management and disease spread. While data sources such as GPS trackers or Call Data Records are temporally-rich, they are expensive, often not publicly available, or garnered only in select locations, restricting their wide use. Conversely, geo-located social media data are publicly and freely available, but present challenges due to their sparse nature. Further, much existing work has focused on predicting next location only, though knowledge of an entire movement timeline is relevant for emerging applications. Accordingly, we propose a stochastic framework, Intermediate Location Computing (ILC) which combines approaches from several existing mobility prediction methods, alongside community behavior, to predict every missing location from an individual's social media timeline. We compare ILC with several state-of-the-art approaches. For three major cities, ILC predicts at 1 and 2-hour resolution with up to 86 show how amount of community data improves prediction, and that community movement improves prediction of an individual's movement more on weekends versus weekdays. Overall this work presents a new algorithm to predict practical and continuous individual-level mobility patterns with sparse but readily available social media data.
READ FULL TEXT