The Optimization Of Ridesharing Matching Based On Attribute Representation Network

Existing work on the ride-matching is based on the spatial structure to enable the optimization.However,Some problems e.g.,the explosion of searching due to the large dimensions and the gap between the information,significantly affect the implementation efficiency.Although network representation has become prominent way to model the information,most of the existing methods often focused on the network structures,and didn’t integrate the individual preferences of the riders,that leads to the problem of low acceptance.It is necessary to adopt the individual attributes of riders within the ride-matching procedure.Thus,we propose an optimization strategy via attribute network representation,considering riders preferences and their travels.The contributions of this paper are summarized as follows: First we extend R-Tree to retrieve vehicle trajectory and riders ’origin-destination(OD)for the candidate set of meeting points between driver and rider.To reduce the scale of points and to accelerate the matching,we filter the points by reducing the time window and distance savings of meeting points.Based on this,we construct an attributed ridesharing network(ARN)based on heterogeneous information network(HIN)which models the distance preference attributes of riders.Then,with ARN,we define a travel meta-path model implying the temporal and spatial semantics.Reducing waiting time is considered as the optimization goal,and we obtain the driver-rider pair vectors in terms of network structure and individual preference respectively.At the same time,one-hot encoding is used to realize the attribute representation of network nodes and generate rider vectors that only consider these attributes.After that we propose a fusion-based approach for ride-matching via attention mechanism.Two kinds of vectors can be adaptive combined with their own weights.At last,the available drivers can be suggested by sorting the similarity of vectors of each driver-rider pair after fusion.We used Didi Gaia opening data in Chengdu to carry out a large number of experimental verifications.The results prove the effectiveness of the proposed method in optimizing matching,and the optimization scheme can provide riders with personalized matching results.Compared with a variety of other methods,our method has achieved a significant improvement in the matching effect.