Sensitivity Analysis of Multi-Level Geohash Fusion in BERT for POI-Based Urban Land Use Classification

To systematically investigate the impact of spatial encoding granularity on model performance, we conduct a sensitivity analysis across multiple geohash levels. By integrating geohash representations of varying resolutions into a BERT-based framework, this experiment evaluates how different geohash fusion strategies influence POI-based urban land use classification. Specifically, geohash levels 5, 6, and 7 are adopted to capture spatial context from coarse to fine scales, enabling a comprehensive assessment of the role of spatial detail in enhancing semantic understanding and classification accuracy.