Research On Ridesharing Matching Scheme Based On Network Embedding

Ridesharing is an emerging mode that enables individuals with matching itineraries and schedules to share a ride in a personal vehicle.By increasing the occupancy of vehicles,the driver and rider(s)typically share the associated costs(e.g.fuel,tolls,parking fees)so that each benefits from the shared ride.The existing ridesharing matching methods are mainly done by similar trajectory clustering or aggregation of a few travel features via linear model.However,the methods didn’t consider the relationship between features,since the features of driver and passenger were independently represented and learned.It thus is difficult for them to represent the complex and sparse semantic relationship between ridesharing participants,further limits the accuracy of matching.In this article,we study the problem of ridesharing matching,which is implemented as a similar participant searching problem under a heterogeneous network.We propose a scheme,RShare Scheme,to exploit heterogeneous networks with multiple types of nodes and links derived from real data on ridesharing behavior,and to embed a ridesharing network in the context of a ridematching task.Specifically,we first formalize two types of ridesharing,end-points and en route,and extract features in terms of user trajectories and sentiment.Based on the features extracted,the ridesharing will be represented by a structural heterogeneous information network(HIN).We then investigate the ride-matching problem of embedding for HIN,which incorporate meta-paths derived from a ridesharing HIN to link the drivers and passengers.Lastly,we allow RShare Scheme to learn a unique embedding representation to encode the travel information on the participants,and evaluates the similarity of participants.The contributions of this article are summarized as follows.(1)Ridesharing classification and feature construction.Firstly,the driver’s trajectory information and the passenger’s annoucements are preprocessed.Then,ridesharing is divided into end-point ridesharing and en route ridesharing,according to the spatial relationship between the passenger’s OD and the driver’s trajectory.The passengers’ sentiment characteristics and associated travel features are identified.Besides,for en route ridesharing,the optimal meeting point search algorithm is proposed to find out the passengers having the lowest travel time cost by picking up at a meeting points.(2)Construction of a ridesharing travel network.Based on the theory of heterogeneous information network,we form a ridesharing travel network that includes different types of objects such as users,location,time,and activities,and object relationships.In such a network,users represent drivers and passengers,and the location is the passenger’s OD or optimal meeting point and activity refers to the type of activity that the user is engaged in at that location.(3)Matching similar participants for ridesharing.We conduct meta-path-based random walks to generate a sequence of neibhbors,and perform a heterogeneous skip-gram over the generated sequences to build a feature vector for each node.We extend the existing metapath2 vec model to rank potential passengers under two types of meta-paths for different ridesharing types.(4)Through extensive experiments,we demonstrate the efficacy of the presented method in ridesharing matching problems.Considering two types of benchmarks,such as network embedding and similar participant searching,compared with the above two methods,obvious improvement has been achieved.