New MESSY STREETS benchmark highlights geocoder performance gaps on real-world addresses

A new benchmark called MESSY STREETS has been introduced to evaluate geocoding systems on real-world, often malformed, addresses. This benchmark, derived from the Web Data Commons corpus and referencing OpenAddresses and OpenStreetMap, specifically tests how geocoders handle addresses with surface-form divergences, missing components, or errors. Results indicate a significant performance gap between commercial and open-source geocoders, with non-canonical address forms contributing substantially to recall loss. AI

IMPACT Highlights the challenges in real-world data processing for AI systems, suggesting improvements in normalization could narrow performance gaps.

RANK_REASON The cluster describes a new academic benchmark paper. [lever_c_demoted from research: ic=1 ai=0.4]

Read on arXiv cs.CL →

• arXiv
• Hugging Face
• MESSY STREETS
• Nominatim
• OpenAddresses
• OpenStreetMap

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →