Inferring routing preferences from user-generated trajectories using a compression criterion

The optimal path between two vertices in a graph depends on the optimization objective, which is often defined as a weighted sum of multiple criteria. When integrating two criteria, their relative importance is expressed with a balance factor alpha. We present a new approach for inferring alpha from trajectories. The core of our approach is a compression algorithm that requires a graph G representing a transportation network, two edge costs modeling routing criteria, and a path P in G representing the trajectory. It yields a mini-mum subsequence S of the sequence of vertices of P and a balance factor alpha, such that the path P can be fully reconstructed from S, G, its edge costs, and alpha. By minimizing the size of S over alpha, we learn the balance factor that corresponds best to the user's routing pref-erences. In an evaluation with crowd-sourced cycling trajectories, we weigh the usage of official signposted cycle routes against other routes. More than 50% of the trajectories can be segmented into five optimal sub-paths or less. Almost 40% of the trajectories indicate that the cyclist is willing to take a detour of 50% over the geodesic shortest path to use an official cycle path.