MTUKG: A Multi-scale Temporal Urban Knowledge Graph Dataset for Knowledge-Enhanced Spatiotemporal Prediction

MTUKG is a multi-scale temporal urban knowledge graph dataset covering New York City and Chicago from January 2014 to December 2025. It organizes urban entities at three spatial scales: micro-level POIs, roads, and junctions; meso-level blocks and functional zones; and macro-level areas and administrative districts. Static spatial and semantic knowledge is represented as triples, while dynamic urban knowledge, including POI events, road repairs, road-network changes, and land-development events, is represented as time-interval facts with explicit start and end dates. Temporal functional zones are constructed quarterly from road-bounded blocks by integrating spatial structure, POI semantics, spatial decay, and dynamic event information. The release provides entity and relation mappings and predefined training, validation, and test sets for both static and temporal facts, supporting temporal knowledge graph reasoning and knowledge-enhanced urban spatiotemporal prediction.