Extracting geographic features from the Internet to automatically build detailed regional gazetteers

Abstract

The utility of every imaginable application which incorporates a gazetteer hinges on the simple fact that the resulting system will only be as useful, complete, or accurate as the underlying gazetteer itself. A major issue confronting gazetteers utilized in systems today is that they are not complete and measures of their accuracy are largely unknown. In this paper we describe a methodology which addresses this problem by automatically generating highly complete and detailed regional gazetteers from Internet sources. We utilize information extraction and integration techniques to automatically obtain geographic features and associated footprints and feature types from freely and widely available online data which could be applied to create a gazetteer for nearly any area. We discuss the distinguishing characteristics of the generated gazetteer and extend previous work to define measures which can be used to assess the completeness and accuracy of gazetteers. Using these measures, the generated gazetteer is evaluated against the Alexandria Digital Library Gazetteer and the Los Angeles Comprehensive Bibliographic Database. Our results indicate that a gazetteer created by our methods will be at least as complete as any gazetteer currently available for certain feature classes, while falling short in others. We conclude by offering suggestions to address these shortcomings.

Automatic gazetteer enrichment with user-geocoded data

“…7 The Fig. 1 An alternative to obtain more precise locations would be to mine addresses with street names, cities and zip codes, as did [11]. However, these addresses will not be rich in vernacular names, as are the sources we use for our study.…”

“…Current methods build or augment gazetteers automatically by mining the web [26], [11], [8]. Others have experimented with gazetteer building by using social network sources [17], [20].…”

Multi-source Toponym Data Integration and Mediation for a Meta-Gazetteer Service

“…Goldberg et al [2] created enriched name and feature type data for merged address (parcel) level places using multiple representations of the same place derived from online residential and commercial phone books. Equivalence of features was established in terms of equivalence of the name and address attributes.…”

“…To provide effective support for these sorts of applications raises the challenge of creating a gazetteer that can maintain access to a wide range of place name terminology relating to many different sorts of features at arbitrary locations. In practice however, because the quality of the content of a gazetteer will depend upon the application for which it is required [2], it is not possible to specify the characteristics of a single ideal gazetteer, except perhaps at a generic level.…”

Generating Named Road Vector Data from Raster Maps

“…These raster maps are easily accessible compared to other geospatial data (e.g., road vector data) and present a unique opportunity for obtaining road information for the areas and time periods where and when road vector data do not otherwise exist. For example, we can generate named road vector data (road vector data that have a road-name attribute) from historical maps and build an accurate geocoder [Goldberg et al, 2009] or a gazetteer for spatiotemporal analysis of human-induced changes in the landscape.…”