Building a global location search service

Abstract

We present a crosslingual location search service that works across multiple countries and deals effectively with ambiguous and ill-formed queries. The system returns a ranked list of spatial regions (points, lines and polygons) that best match users' text queries, which can range from postal addresses to unstructured queries that list a few col-located map features. The system's robustness comes from a novel approach that exploits spatial coherence to identify viable interpretations of input text. Unlike existing state of the art geocoding systems, our system requires no region-specific rules, training or customization, and thus may be built to cover any region for which detailed map data is available, making it possible, for the first time, to rapidly build location search services for new regions, and more generally, approximate text search over arbitrary spatial repositories. Our system has been shown to outperform commercial geocoding systems, especially when spelling or format variations are introduced. We demonstrate our sys-tem's capabilities by showing results for a variety of text queries over a large dataset from several countries with widely differing address formats.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 2 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.