Joint Estimation and Prediction of City-wide Delivery Demand: A Large Language Model Empowered Graph-based Learning Approach

2024-08-30

Transportation Research Part E: Logistics and Transportation Review

AI Summary

To address the challenges of dynamically modeling urban-scale on-demand delivery demand and enabling cross-city generalization, this paper proposes an LLM-enhanced spatiotemporal graph neural network framework. Methodologically, it introduces the first coupling of a fine-tuned LLaMA-2 large language model with a graph attention network (GAT), integrating unstructured urban semantic information—including POI descriptions and policy texts—to construct transferable dynamic graph topologies; it further designs a spatiotemporal encoder and a multi-task joint optimization loss function. The key contribution lies in overcoming traditional time-series models’ reliance on static graph structures and isolated features, enabling end-to-end joint modeling of demand estimation and multi-step forecasting. Evaluated on three real-world megacity datasets, the framework reduces demand estimation error by 23.6% and achieves a 17.4% improvement in 72-hour forecast MAE over state-of-the-art methods, significantly enhancing fleet dispatch responsiveness and cross-city generalization capability.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Problem

• Joint estimation and prediction of city-wide delivery demand
• Generalization of delivery demand models to new cities
• Integration of large language models for geospatial knowledge

Innovation

• Graph-based spatiotemporal learning for demand estimation
• LLM embeddings extract geospatial knowledge encoding
• Transferable model enhances cross-city generalization