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

2026-06-12 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Geospatial AI · Depth: Expert, long

Summary

PlaceFM is a novel geospatial foundation model introduced in 2025 that generates general-purpose place embeddings from large-scale Point-of-Interest (POI) data using a training-free graph condensation method. It addresses the limitation of existing models that struggle to reason about context-rich "places" across different spatial granularities. PlaceFM condenses a nationwide POI graph, comprising approximately 433K POIs integrated from Foursquare and OpenStreetMap data in the U.S., to create adaptable embeddings. The methodology involves constructing a k-nearest neighbor (k-NN) POI graph, followed by a two-stage training-free graph condensation: Feature Propagation, which aggregates neighborhood information, and Representation Clustering, which uses k-means to group POIs into semantically meaningful places at user-defined granularities such as zip code, city, or state. Experiments demonstrated K-Means' superior clustering performance over K-Medoids and identified a k-NN value of 5 as optimal for feature propagation.

Key takeaway

For geospatial data scientists developing location-aware applications, PlaceFM offers a scalable approach to generate flexible, multi-granular place embeddings without extensive pretraining. You can integrate its training-free graph condensation method into your data pipelines. This captures context-rich regions beyond fixed boundaries, enabling more nuanced spatial analysis. Adapt representations to your specific analytical needs for tasks like urban planning or anomaly detection.

Key insights

PlaceFM generates multi-granular, training-free place embeddings from POI graphs using graph condensation, enhancing geospatial intelligence.

Principles

• Graph condensation distills large graphs.
• Multi-hop aggregation enriches POI context.
• Clustering forms semantically coherent places.

Method

PlaceFM constructs a k-NN POI graph from F-OSM data, then applies training-free graph condensation via feature propagation and k-means clustering to generate multi-granular place embeddings.

In practice

• Integrate Foursquare and OpenStreetMap data.
• Use k-NN (k=5) for POI graph edges.
• Apply k-means for place clustering.

Topics

• Geospatial Foundation Models
• Place Representation Learning
• Point-of-Interest Data
• Graph Condensation
• Foursquare OpenStreetMap
• Multi-scale Geospatial Analysis

Code references

• onspatial/f-osm
• Emory-Melody/GraphSlim

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• A Closed-Form Persistence-Landmark Pipeline for Certified Point-Cloud and Graph Classification — PLACE offers a closed-form pipeline for point cloud and graph classification with certified guarantees from persistent homology.
• CausalPOI: Spatio-Temporal Graph-Based Causal Modeling for Cold-Start POI Check-in Forecasting — CausalPOI uses spatio-temporal causal modeling to predict check-ins for new POIs, outperforming correlation-based methods.
• Spatial Representation Learning Beyond Pixels: Unifying Raster Data and Vector Semantics for Human-Centric Geospatial Foundation Models — Unifying raster and vector geospatial data is essential for developing human-centric, semantically grounded Earth Observation Foundation Models.
• Efficient frequent directions algorithms for approximate decomposition of matrices and higher-order tensors — New frequent directions algorithms enhance low-rank matrix and tensor approximations for efficiency and accuracy.
• GeoGNN: Time Series Geo-Localization using Two-Tower Graph Neural Networks — GeoGNN's two-tower GNN infers time series geographic origins, boosting accuracy by integrating spatial and temporal embeddings.
• FlatVPR: Plug-and-play Geo-linear Residual Adapter for Geometric Rectification of Foundation Model Feature Manifolds — FlatVPR flattens foundation model feature manifolds for accurate VPR with sparse anchors via a learnable residual adapter and geometric loss.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.