Multi-View Hypergraph-Based Ride-Sourcing Origin-Destination Demand Prediction

Ride-sourcing services have become a vital urban mobility mode, offering flexibility and convenience. Accurate origin-destination (OD) demand prediction is crucial for ride-sourcing platforms to deliver efficient on-demand mobility services. Existing studies have tried to extract complex spatial-temporal dependencies of OD pairs from different aspects. Differently, we propose a novel multi-view hypergraph OD demand prediction framework, which utilizes hypergraphs to associate multiple OD pairs from static and dynamic views simultaneously. In the static hypergraph view, we explore spatial-temporal correlations from three perspectives: points of interest, time series, and motifs. Among them, motifs can capture local higher-order network structures in the OD network. In the dynamic view, we utilize real-time OD demand as prior information to learn representations of multiple OD pairs directly and capture evolving spatial-temporal dependencies. Moreover, a gate mechanism and probability constraints are designed to enhance accuracy and reliability. Finally, experimental results on two large-scale ride-sourcing datasets from Hangzhou and Ningbo verify the superiority of our proposed model over seven baseline models. Accurate OD demand prediction aids in guiding vehicle dispatching and dynamic pricing for ride-sourcing platforms.