Intertemporal Community Detection in Human Mobility Networks

We introduce a community detection method that finds clusters in network time-series by introducing an algorithm that finds significantly interconnected nodes across time. These connections are either increasing, decreasing, or constant over time. Significance of nodal connectivity within a set is judged using the Weighted Configuration Null Model at each time-point, then a novel significance-testing scheme is used to assess connectivity at all time points and the direction of its time-trend. We apply this method to bikeshare networks in New York City and Chicago and taxicab pickups and dropoffs in New York to find and illustrate patterns in human mobility in urban zones. Results show stark geographical patterns in clusters that are growing and declining in relative usage across time and potentially elucidate latent economic or demographic trends.