JointMatch Unifies Request Pairing and Vehicle Assignment via GNNs

JointMatch replaces sequential matching heuristics with a single heterogeneous graph neural network that handles both request bundling and vehicle assignment. By sparsifying the graph via spatial proximity, the framework achieves linear scaling and outperforms the Blossom heuristic on NYC Yellow Taxi datasets.