Recommending routes in the context of bicycling: algorithms, evaluation, and the value of personalization

Users have come to rely on automated route finding services for driving, public transit, walking, and bicycling. Current state of the art route finding algorithms typically rely on objective factors like time and distance; they do not consider subjective preferences that also influence route quality. This paper addresses that need. We introduce a new framework for evaluating edge rating prediction techniques in transportation networks and use it to explore ten families of prediction algorithms in Cyclopath, a geographic wiki that provides route finding services for bicyclists. Overall, we find that personalized algorithms predict more accurately than non-personalized ones, and we identify two algorithms with low error and excellent coverage, one of which is simple enough to be implemented in thin clients like web browsers. These results suggest that routing systems can generate better routes by collecting and analyzing users' subjective preferences.