Spatial RAG for Big Data-Driven urban itinerary recommendation

Urban mobility and tourism rely on heterogeneous Big Data, yet traditional recommender systems often rely on static shortest-path models that ignore complex, dynamic urban indicators. While LLMs offer reasoning capabilities that can mitigate these issues, they lack factual accuracy and are prone to spatial hallucinations. We present UrbanRAG, a Spatial Retrieval-Augmented Generation (RAG) framework that fuses LLMs with neural information retrieval to leverage large-scale urban Big Data. UrbanRAG enables natural-language interaction to generate personalized itineraries by integrating worldwide map data, environmental indicators, and other semantic information. Unlike localized databases, our open-world architecture ensures scalability and portability. Preliminary empirical results demonstrate that UrbanRAG significantly improves spatial grounding and accuracy over closed-book LLMs. By anchoring generative models in real-world spatial data, our framework provides a robust, adaptive solution for data-intensive urban services.