Urban Activity Mining Framework for Ride Sharing Systems Based on Vehicular Social Networks

Abstract

Ride sharing has been widely studied over the past several decades as a means of reducing traffic and pollution by utilizing empty car seats in vehicles that are being driven no matter what. As they increase in popularity, ride sharing applications have already encountered several challenges: Vehicle allocation, price strategy, and route planning, are just a few such examples among many. Tracking human activity patterns with regard to ride share applications can potentially improve these systems in numerous ways. For example, taxi GPS trajectories offer a remarkable source of data for determining human activity patterns, among other things, in cities across the world. However, existing studies either focus solely on meeting order requirements or analyzing Points Of Interest (POI) based only on static information. The former issue cannot solve problems with balancing vehicle allocation, while the latter cannot precisely describe the POI locational feature. In order to develop a more specific analysis of activity patterns for ride sharing systems, we propose a Vehicular Social Network Based Analytical Framework (NBAF) to determine the specific urban activity of ride sharing systems at a low computational cost. The analytical framework contains two special contributions: Firstly, a novel trip mapping method named Trip-Embedding POI Decomposition Method (TEPID) is proposed to describe the feature of geo-nodes in the network. Secondly, the particular features for ride sharing systems are generated by vehicular social networks. Based on this framework, we propose a clustering method to reveal trip activity and regional features. As a case study, we analyze 30 days of taxi trips in New York City in 2016. The results demonstrate that NBAF can effectively determine urban activity and location patterns for vehicle allocation, price strategy, and route planning for ride sharing systems.