Reproduce “Spatio-Temporal Demand Prediction for Food Delivery Using Attention-Driven Graph Neural Networks” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Rabia Latief Bhat, Iqra Altaf Gillani

This research introduces an attention-driven Graph Neural Network (GNN) framework for spatio-temporal demand prediction in food delivery, forecasting both overall demand and Origin-Destination (OD) patterns. Evaluated on a real-world Meituan dataset, the framework achieved a Mean Absolute Percentage Error (MAPE-0) of 0.1972 for demand prediction, outperforming established baselines.

Key insights

  • Optimal spatial granularity (2.5 km grid cells) and temporal granularity (15-minute time slots) are crucial for balancing predictive accuracy and computational efficiency in food delivery demand forecasting.
  • The Linear Recurrence Layer, which captures consistent demand patterns from the same hour across past days, proved vital for prediction accuracy, highlighting the importance of stable, repetitive user behavior.
  • Explicitly modeling Origin-Destination (OD) flows and dynamically weighing neighbor influence through attention mechanisms are critical for enhancing prediction accuracy beyond aggregate demand and uniform spatial weighting.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.