Inferring the Travel Purposes of Passenger Groups for Better Understanding of Passengers

People usually travel to the same destinations and for the same purposes together with other people in groups. Inferring the travel purposes of passenger groups is a very interesting research problem in the field of passenger transport, because it can help us to better understand passengers and should bring about meaningful changes for personalized travel service and decision making of passenger carriers, organizations, and even governments. In this paper, we attempt to solve this problem by utilizing the historical travel records of passengers. To overcome the constraint of the independent and identical distribution assumption of traditional classifiers, we propose a novel iterative classification approach based on the idea of collective inference. First, we construct cotravel networks by extracting social relations between passengers from their historical travel records that are available in carriers´ passenger information systems. Then, we generate a series of sophisticated features for each passenger group in the context of cotravel networks to capture the link structure information between passengers and use the overlapping relations between passenger groups to capture the probabilistic dependence relations between their labels. Finally, we collectively infer the labels of all the groups in an iterative way. Experimental results on a real data set of passenger travel records in the field of civil aviation demonstrate that our proposed iterative classification approach can efficiently infer the travel purposes of passenger groups.