A Multi-strategy Approach to Geo-Entity Recognition

Abstract

Geographic location or place information has become an increasingly integrated and important element in web and online interaction, which is evident in the increasing sophistication and adoption of online mapping, navigational GPS, and location-aware search. A significant proportion of online location context, however, remains implicit in primarily unstructured document text. In order to leverage this location context, such references need to be extracted into structured knowledge elements defining place. A variety of “named entity” extraction methods have been developed in order to identify unstructured location references, alongside other references such as for persons or organizations, but geographic entity extraction remains an open problem. This chapter examines a multi-strategy approach to improving the quality of geo-entity extraction. The implemented experimental framework is targeted for web data, and it provides a comparative evaluation of individual approaches and parameterizations of our multi-strategy method. Results show that the multi-strategy approach provides a significant benefit in terms of accuracy, domain independence, and adaptability.