Structured toponym resolution using combined hierarchical place categories

Abstract

Determining geographic interpretations for place names, or toponyms, involves resolving multiple types of ambiguity. Place names commonly occur within lists and data tables, whose authors frequently omit qualifications (such as city or state containers) for place names because they expect the meaning of individual place names to be obvious from context. We present a novel technique for place name disambiguation (also known as toponym resolution) that uses Bayesian inference to assign categories to lists or tables containing place names, and then interprets individual toponyms based on the most likely category assignments. The categories are defined as nodes in hierarchies along three orthogonal dimensions: place types (e.g., cities, capitals, rivers, etc.), geographic containers, and prominence (e.g., based on population).

Itinerary retrieval

“…based on Combined Hierarchical Place Categories (CHPC) [3], as mentioned in Section 2. The key step in the method is identifying a common thread that can be used to categorize the interpretations of all toponyms in the set (e.g., "cities in Bavaria with population > 10,000").…”

“…The more specific problem of geotagging data tables has been addressed in some settings, such as for ontology extraction [11], entity discovery in Fusion Tables [24], and general spreadsheet and table geotagging [20]. For itinerary geotagging, we use our probabilistic model for geotagging collections of place names [1,3,18], which identifies place categories for geographic table columns, then disambiguates toponyms in the context of that category. The method achieves high accuracy on sample tables and appears to be a good fit for itineraries, which tend to visit places that share similarities of geography, type, and/or prominence.…”

Reading news with maps by exploiting spatial synonyms

“…Another example of this strategy is when a query involves lists of locations, in which case NewsStand tries to use proximity, sibling, and prominence clues to resolve the ambiguity. 1,21 Evaluation. To see how well NewsStand's geotagging performs, rather than display a news category icon at a location, NewsStand can display the actual name of the location by setting the "layers" parameter to "location" instead of to "icon."…”

Spatio-temporal disease tracking using news articles

“…The Web-a-Where system of Amitay et al [6] uses a variety of context features to interpret geographic references in documents and demonstrates the need for incorporating such features in order to geotag text accurately. Unlike many health mapping systems that rely on machinespecified location information or straightforward geotagging procedures, NewsStand incorporates many context features, including the presence of proximity, sibling, and prominence clues [2,5,27], to increase the accuracy of the extracted location references. News Rover [21] is another system, similar to NewsStand, that does aggregation of news videos, rather than articles, for exploration and querying.…”