Exploring human activity behavior and mobility data in carpooling

Abstract

The analysis of human movements has been the subject of several studies since the 70s. In recent years, the exponential growth of location aware devices must allow the study of the behavior of the individuals’ mobility from their trajectories collected. However, a significant part of the available literature is focused on the development of techniques for analyzing trajectories of people from a purely geometric point of view, while a smaller part, but increasingly group is looking at the semantic aspects of mobility. This work presents a contribution to the latest trend, and is concerned with the definition of semantic regularity profiles and the applicability of these concepts to the practice of carpooling. We propose a semantic regularity profile based on the entropy of the spatial and temporal frequency of visits to certain categories of places. We analyze the user’s behavior with respect to regularity and irregularity, identifying users who are more or less loyal to certain locations, in contrast to the irregularity of visiting different places. In a different point of view, an analysis over the place perspective was also performed. A web tool was developed to show on map, for each place of a given category, the computed information about the loyal behavior of their visitors. From the study about regularity, we have evidences that some human activities are not strictly associated to a unique POI (Point of Interest) and neither to a specific schedule of the day. Bringing to the carpooling context, in some situations it is worth for a person to change his destination or the time to perform an activity if there is a possibility of ride for him due all the benefits involved with the carpooling practice. This dissertation also presents a novel matching method for carpooling that is oriented to the passenger’s intended activity, aiming to boost the possibilities of rides. Three algorithms for carpool matching are proposed, which manipulates differently the spatial and temporal dimensions. Using a real data set of trajectories, we conducted experiments and our results showed that the proposed matching algorithms improved the traditional carpooling approach in +46.84% when the spatial dimension was considered, in +50.89% when the temporal dimension was prioritized and in +82.30% when both dimensions were tackled.