SEXTANT: A Computational Framework for Scalable and Efficient Correlation of Spatio-Temporal Trajectories

Discovering pairs of related entities based on their spatio-temporal behavior has applications of scientific, commercial, and public interest. For example, inferring relationships between animals based on data collected from tracking devices can help biologists better understand group behavior and family structures; identifying commonalities in customer behavior can enable better targeted marketing; and discovering correlations in the movement patterns of maritime vessels can help crack down on illegal fishing or smuggling activity. However, this task becomes especially challenging when many entities are being observed simultaneously and when observations of different entities are made asynchronously and at different frequencies. We propose SEXTANT, a computational framework for performing spatio-temporal correlation at scale, implemented using efficient data structures over a distributed architecture. SEXTANT's core capability consists of algorithms that, given a large set of time-stamped and geo-located observations, identify pairs of trajectories with high spatial proximity over time, demonstrating shared pattern-of-life behavior. Experiments on real-world and synthetic datasets demonstrate SEXTANT's ability to efficiently capture correlated behavior at scale while being robust to asynchronicity, sparsity, and heterogeneity in the data, out-performing existing methods.