Urban Trajectory Analytics: Day-of-Week Movement Pattern Mining Using Tensor Factorization

Abstract

This paper demonstrates the use of non-negative tensor factorization to extract underlying spatio-temporal movement patterns from large-scale urban trajectory data. Individual trajectory data obtained from public transport smart card systems and roadside Bluetooth detectors are represented as a dynamic graph of region-to-region flows to obtain structured data describing flow interactions between regions across time-of-day and day-of-week. Tensor factorization is then applied to these dynamic graphs to characterize traveler movements on different days of the week. The results unveil distinct day-of-week patterns in public transport passenger and roadway vehicle movements, providing insight into the diverse aspects of urban mobility.