Determining the bounds of geographic regions is an important task for geographic search engines which use concept@location-type of queries. The location a user specifies is often not contained in the underlying gazetteer or geographic database, which might be due to vernacular descriptions of regions or because the location is not a geographic region in the narrow sense, which is the case in queries like campground near theme park. In the present paper we describe different ways for automatically determining a geographic footprint for those locations so that a geographic search engine is able to deal with all kinds of location-descriptions. The same approaches can be used to visualize the geographic correlation of arbitrary terms, like the visualization of the spread of certain colloquialisms.
The basic idea is to mine locations found in the top documents resulting from a query consisting of the terms the user has chosen to specify the location. We describe how this can be done using kernel density estimation, clustering and a combination thereof.