146 — Discovering the hidden community structure of public transportation networks

When a city uses smart cards in order to access public transportation, it opens up huge possibilities for data-aggregation and tracking: These cards make it possible to track individual rides around a city, learning about trends in transportation as well as — sorta creepily — the particular ride patterns of an individual. This sort of dataset lends itself nicely to graph theory, and so Hajdu et al perform a few analyses on an example dataset of Twin Cities, MN.

The authors propose two graph transformations: a contact network and a transfer network — networks that shows the “community-ness” between two individuals. For example, if you and I ride through the subway station together, we are more related than two individuals who ride six hours apart. Likewise, if we regularly leave from the same station at the same time of day, we’re likely to have a very high measure of correlation using this model.

A conveninent way to imagine this weighting — and, the authors note, a very good application for this technology — is modelling the spread of infectious disease over a geography where there are complex but consistent transportation patterns (such as subways or buses).

These community networks enable city-level optimization as well: The graph makes common combined trips easy to find: For example, if 90% of travelers from $A$ to $B$ continue on to point $C$, then perhaps it makes sense for the city to invest in a direct line from $A$ to $C$.