PlaceRep: Geospatial Place Representation Learning from Large-Scale Point-of-Interest Data

PlaceRep is a novel geospatial representation learning method that clusters spatially and semantically related Points of Interest (POIs) to create place-level embeddings, moving beyond traditional administrative boundaries. By leveraging large-scale POI data from U.S. Foursquare, PlaceRep eliminates the need for pre-training, enabling efficient multi-granular geospatial analysis and achieving significant improvements in tasks like population density estimation and housing price prediction, with up to a 100x speedup in generating region-level representations. This advancement is crucial for practitioners seeking scalable solutions in urban analytics and geospatial modeling.