Crowd-sourced carpool recommendation based on simple and efficient trajectory grouping

We propose a novel carpool recommendation method that is based on simplifying a user's movement traces. An effective carpool recommendation system requires that users following the most similar driving routes be identified and that these routes then be consolidated into one or more 'recommended' optimal carpool driving route options that users' can choose from. Currently mobile users generate a high volume of detailed trajectory data, making it difficult to efficiently derive optimal recommendations. We devise a simple method for building a user's trajectory profile, which is then used in deriving the recommendation(s). Unlike an origin-destination based analysis, which matches up riders with drivers, our method creates feature points along a simplified path that has been derived from the mobile user's moving trace. This maintains the sequence of movements and preserves feature points, including intersections and common places. Feature points are mapped using quad-keys as part of a tile map system that enables a membership of feature points within the range of a given area. Using this membership, recommendations for optimal carpool routes are made by measuring how users share common quad-keys along their trajectories. We tested our proposed method using historical traces of two crowd-sourced projects: TrafficPulse and GeoLife. The results show the advantage of the proposed method for dealing with a high volume of detailed mobile trajectory data, both in terms of requiring reduced data storage space and requiring reduced computational cost.