Personalized public transport mobility service: a journey ranking approach for route guidance

Abstract

Abstract The use of smartphone applications (apps) to acquire real time and readily available journey planning information is becoming an instinctive behavior of public transport (PT) users. Through the apps, a traveler not only seeks a path from origin to destination, but also a satisfactory path that caters to the traveler’s preferences at the requested time of travel. In other words, to strive for a personalized PT service. The personal preferences are naturally enabled because of the existence of multiple attributes associated with alternative PT routes. For instance, preferences can be connected to attributes of time, cost and convenience. Initially this work establishes an adjusted design framework, to those existed nowadays, for a personalized PT service by integrating users’ experiences using apps with operators’ data sources and operations modeling. The work then focuses on its key component, namely the personalized route guidance methodology. In addition to using the classic shortest path method, two developments are suggested: a k-weighted shortest path method and a novel lexicographical shortest path method with a just noticeable difference (JND) consideration. The latter adopts lexicographical ordering to capture traveler preferences over different PT attributes following Ernst Weber’s law of human perception threshold. However, Weber’s law violates the axiom of transitivity required for an implementable algorithm, and thus a revised method is developed with correctness proven algorithms for ranking different paths. A small example is used as an explanatory device to illustrate the differences between the three route-guidance methods and to demonstrate the effects of the JND perception threshold on the order of the alternative PT routes. A simulation study was conducted using the Copenhagen PT network. The results show that the average reduction of the value of the most important PT traveler’s attribute is 12.3% for the k-weighted shortest path method, and 13.4% for the lexicographical JND-based shortest path method in comparison with the classical shortest path method. The computation time indicates favorable potential for real life applications.