Modeling Urban Mobility Networks Using Constrained Labeled Sequences.

Models of urban mobility patterns are key inputs to urban analytics, e.g., planning the transportation infrastructure. Typically, only limited data about mobility is available, due to privacy and data collection challenges. Here, we study the problem of reconstructing mobility traces for all individuals from flow measurements on transportation network links. We formalize this as the constrained label sequence problem (denoted by CLS). CLS is the problem of constructing labeled sequences with constraints that correspond to flows. The proposed formulation is quite general and allows us to consider bounds on the total flow through a set of links, instead of an individual link. We show that CLS is computationally hard to solve exactly, and design an algorithm with rigorous approximation guarantees. We also study the complexity of counting and sampling from the set of mobility traces consistent with such measurements. Finally, we demonstrate our results using transportation flows across zones around Washington DC and data from the American Community Survey Commuting Flows.