Extracting metadata for spatially-aware information retrieval on the internet

Abstract

This paper presents methods used to extract geospatial information from web pages for use in SPIRIT, a new Geographic Information Retrieval (GIR) system for the web. The resulting geospatial markup tools have been used to annotate around 900,000 web pages taken from a 1TB web crawl, focused on regions in the UK, France, Germany and Switzerland. This paper discusses a versatile geo-parsing tool for extracting spatial metadata based upon the GATE Information Extraction (IE) system, and a simple geo-coding program based on default sense to assign spatial coordinates to extracted locations. A preliminary analysis of markup accuracy for geo-parsing and geo-coding is provided, and an initial statistical and geographical analysis of the SPIRIT collection presented.

The design and implementation of SPIRIT: a spatially aware search engine for information retrieval on the Internet

“…The primary reason is simple -the number of georeferenced documents is relatively small -the SPIRIT collection for the whole of the UK, for example, consists of only around 340,000 documents. Further, as discussed in Clough (2005) not all documents are correctly georeferenced (around 90% in the experiments reported by Clough).…”

“…We used a default sense approach and global geographical world knowledge to resolve ambiguity based on features from the geographical resources available to us. Again, based on the evaluation described in Clough (2005), we were able to ground correctly around 89% of all place names. Locations were assigned an appropriate bounding box, representing a spatial extent derived from polygonal data stored in the geo-ontology, since the overheads associated with passing polygons through the system were too high.…”

“…A geographical ontology acts as a repository of knowledge about place names, and relationships between them, for the regions covered by the search engine. Data in this repository are used to recognise the presence of place names in web resources and hence to "ground" the web resources to geographic locations, a process referred to as geo-tagging (Clough, 2005). The geographic ontology is also used to disambiguate place names in a user query and to generate a query footprint that reflects the region of space to which the query refers.…”

“…Locations include regions such as villages, towns, cities, counties and places of interest represented spatially as points and polygons. Based on empirical evaluation (Clough, 2005), our approach using gazetteer lookup with additional context rules gave an accuracy of around 72% and 25% false positives for all annotations found.…”

Visualising the south yorkshire floods of '07

“…The General Architecture for Text Engineering (GATE) system [5] was used to extract geographical information (as used in previous work [6]). The Ordnance Survey (OS) 50k Gazetteer and Locator resources were used to assist extraction of locations and assignment of spatial coordinates.…”