POIFormer: A Transformer-Based Framework for Accurate Point-of-Interest Attribution

Area of Science:

• Data Science
• Urban Computing
• Mobility Analytics

Background:

• Accurate Point of Interest (POI) attribution is crucial for urban planning and personalized services.
• Existing methods struggle with GPS inaccuracies and dense urban POI clustering.
• Proximity alone is insufficient for reliable POI attribution.

Purpose of the Study:

• To introduce POIFormer, a novel Transformer-based framework for accurate and efficient POI attribution.
• To address the limitations of existing approaches by jointly modeling multiple data dimensions.
• To develop a practical and generalizable solution for real-world mobility datasets.

Main Methods:

• Utilized a Transformer architecture with self-attention to model complex interactions between spatiotemporal, contextual, and behavioral features.
• Incorporated user visit history (past and future, with current visit masked) and aggregated crowd behavior patterns (KDEs).
• Developed a framework that integrates spatial proximity, visit timing/duration, POI semantics, and user mobility patterns.

Main Results:

• POIFormer demonstrated significant improvements in POI attribution accuracy over existing baselines.
• The model excelled in challenging real-world scenarios with spatial noise and dense POI clusters.
• Achieved accurate and efficient attribution in large-scale, noisy mobility datasets.

Conclusions:

• POIFormer offers a robust and scalable solution for accurate POI attribution in diverse urban environments.
• The framework's ability to generalize across data sources and contexts enhances its practical applicability.
• This approach overcomes key challenges in mobility analytics, paving the way for improved urban insights.