Mining Influential Terms for Toponym Recognition and Resolution

Abstract

The detection of toponyms present in text has appeared as a useful resource for many diff erent applications, such as for social network analysers and for geographic search engines. The variety of ambiguities present in the geoparsing process represents one of the main challenges related to the process of detecting toponyms, bringing the need for treating this problem with careful attention. One important technique to detect toponyms is based on the presence of infl uential terms, which are terms that could indicate the existence of geographical references in the text. This paper presents an approach to automatically identifying relevant infl uential terms for a given language, as well as a set of attributes relating these terms with toponyms. The technique presented here was validated with an existing geoparser, using a training set based on online news. The results indicate the technique is eff ective in identifying infl uential terms, and has shown that the geoparser's capabilities of detecting toponyms have improved by using the generated list of infl uential terms.