OD-Prophet: Toward Efficiently Predicting Individual Origin-Destination Travel Demand in Location-Based Services

Human mobility prediction plays a critical role in various location-based services. Different from existing works on crowd-level mobility prediction and individual-level next POI or destination prediction, this is the first work to investigate the individual origin-destination (OD) prediction to enable novel and personalized services. Specifically, based on two complementary data sources, i.e., mobile application location data with sparse GPS trajectories and public transport payment data with OD information from a popular social network platform, we aim to simultaneously predict the origin and destination of a new trip. However, we face three unique challenges: 1) uneven and sparse location sampling; 2) time shifting of periodical mobility, and 3) personalized travel patterns. To this end, we propose OD-Prophet, a contrastive learning-based multiview attention network, consisting of 1) a multiview embedding component to obtain personalized representations with enhanced group information; 2) a correlation-enhanced hybrid attention mechanism to alleviate the time shifting of periodical mobility; and 3) a supervised contrastive learning framework to generate high-quality representations from sparse trajectories and improve the generalization capability. Extensive evaluations on two real-world mobility data sets demonstrate that OD-Prophet outperforms the SOTA baselines by more than 12.8% accuracy.