Scholarlink

<p>GeoJitter is an open-source Python package for region-aware randomization of node locations in spatial networks, designed to preserve network structure while mitigating privacy risks and supporting spatial visualization of incomplete or uncertain data. Existing open-source randomization techniques rarely incorporate geospatial boundaries, limiting their applicability for spatial analysis; GeoJitter addresses this gap and provides tiling solutions when region data is unavailable. Performance was compared against Radius and K-Nearest Neighbor randomization across 100 trials on Brightkite and Gowalla networks; structural metrics remained stable with the most disparity in average path length, which changed by 0.14–0.28 (versus 0.06–0.09 for benchmarks). Changes in betweenness were near zero (− 0.0001 overall), clustering coefficient shifted by approximately 0.009, and modularity by − 0.015. These results demonstrate comparability to established methods while introducing a flexible, region-aware approach for privacy-sensitive spatial analysis.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

GeoJitter: a flexible toolkit for region-aware node jittering in spatial networks

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月28日

查看全文
  • Nina Fiore,
  • Sebastian Neumann,
  • Geoffrey Moores

摘要

GeoJitter is an open-source Python package for region-aware randomization of node locations in spatial networks, designed to preserve network structure while mitigating privacy risks and supporting spatial visualization of incomplete or uncertain data. Existing open-source randomization techniques rarely incorporate geospatial boundaries, limiting their applicability for spatial analysis; GeoJitter addresses this gap and provides tiling solutions when region data is unavailable. Performance was compared against Radius and K-Nearest Neighbor randomization across 100 trials on Brightkite and Gowalla networks; structural metrics remained stable with the most disparity in average path length, which changed by 0.14–0.28 (versus 0.06–0.09 for benchmarks). Changes in betweenness were near zero (− 0.0001 overall), clustering coefficient shifted by approximately 0.009, and modularity by − 0.015. These results demonstrate comparability to established methods while introducing a flexible, region-aware approach for privacy-sensitive spatial analysis.