Assessing the Certainty of Locations Produced by an Address Geocoding System

Abstract

Addresses are the most common georeferencing resource people use to communicate to others a location within a city. Urban GIS applications that receive data directly from citizens, or from legacy information systems, need to be able to quickly and efficiently obtain a spatial location from addresses. In this paper we understand addresses in a broader perspective, in which not only the conventional elements of postal addresses are considered, but other kinds of direct or indirect references to places, such as building names, postal codes, or telephone area codes, which are also valuable as locators to urban places. This broader view on addresses allows us to work with two perspectives. First, in the ontological definition, modeling, and implementation of an addressing database that is flexible enough to accommodate the variety of concepts and address formats used worldwide, along with direct and indirect references to places. Second, in the definition of an indicator that is able to quantify the degree of certainty that could be reached when a user-given, semi-structured address is geocoded into a spatial position, as a function of the type and completeness of the available addressing data and of the geocoding method that has been employed. This indicator, which we call Geocoding Certainty Indicator (GCI), can be used as a threshold, beyond which the geocoded event should be left out of any statistical analysis, or as a weight that allows spatial analysis methods to reduce the influence of events that have been less reliably located. In order to support geocoding activities and the determination of the GCI, we propose a conceptual schema for addressing databases. The schema is flexible enough to accommodate a variety of addressing systems, at various levels of detail, and in different countries. Our intention is to depart from the usual geocoding strategy employed in commercial GIS products, which is usually limited to the average American or British address format. The schema also extends the notion of

Ontology-driven discovery of geospatial evidence in web pages

“…The quality of geocoding varies according to the strategy that is employed [49], and in turn the strategy depends on the quality and type of available addressing data. Such data may be found as numbering ranges over street centerline segments (also known as street geocoding, based on the TIGER file approach [47]), as numbers associated to land parcel centroids [35], or as numbers associated to individual buildings [11]. The last strategy has been developed as a response to addressing problems that are common in large cities of emerging countries, such as Brazil, due to the common occurrence of problems such as ambiguous street names and irregular numbering, which often rend commercial geocoding software useless.…”

“…The approach is based on an ontology that has been designed to facilitate the process of recognizing, extracting, and geocoding partial or complete references to places embedded in text. It combines the extraction ontology approach described earlier with advances on urban gazetteers [42] created from data available on the Web and geocoding techniques [11]. This ontology, called OnLocus, is described in Section 4.…”

“…Addresses incorporate a time-proven system for locating a place based on a formal description. Addresses are practically universal in urban areas and embed the idea of narrowing down the search for a given place using a hierarchy of information items: country, state or province, city, neighborhood, street, building number [11]. Since postal addresses are usually presented in a standardized way, it is possible to recognize them in text from a previously determined template.…”

“…Variations in Web page address elements, such as abbreviations, punctuation, and line breaks, complicate creating an address parser [39]. There may have some parts missing, such as the country [11,31]. In such cases, indirect references serve as additional evidence to determine or to infer the missing pieces.…”

“…If neither an exact nor an approximate location can be determined, a general location can often be established using information such as the neighborhood name, postal code, or city limits [11].…”