A survey on the geographic scope of textual documents

Highlights

• A structured and comprehensive view on contributions to the problem of determining the geographic scope of documents.

• Review of relevant definitions and concepts.

• A discussion of the main techniques currently used to address the Geographic Scope. Resolution (GSR) problem and its subproblems.

• A proposal for a future research agenda in the area.

Abstract

Recognizing references to places in texts is needed in many applications, such as search engines, location-based social media and document classification. In this paper we present a survey of methods and techniques for the recognition and identification of places referenced in texts. We discuss concepts and terminology, and propose a classification of the solutions given in the literature. We introduce a definition of the Geographic Scope Resolution (GSR) problem, dividing it in three steps: geoparsing, reference resolution, and grounding references. Solutions to the first two steps are organized according to the method used, and solutions to the third step are organized according to the type of output produced. We found that it is difficult to compare existing solutions directly to one another, because they often create their own benchmarking data, targeted to their own problem.

Introduction

The demand for geographic data in applications on the Web is increasing. One of the most important resources to support this increased interest is the ability to recognize references to places in Web documents. If documents can be correctly and efficiently linked to places mentioned directly or indirectly in them, it becomes possible to improve and innovate in directions such as geographic indexing and querying, finding relationships based on spatial proximity or containment, and detecting localized trends for events and phenomena mentioned in social media.

A large share of the information available on the Web is geographically specific (Delboni et al., 2007, Vaid et al., 2005, Vasardani et al., 2013). References to geographic locations appear in the form of place names, postal addresses, postcodes, historical dates, demonyms, ethnicity, typical food and others. Many queries include place names and other geographic terms (Delboni et al., 2007, Sanderson and Kohler, 2005, Silva et al., 2006). Therefore, there is demand for mechanisms to search for documents both thematically (for instance, using a set of keywords) and geographically, based on places mentioned or referenced by the text (Zong et al., 2005). Similar techniques and resources can also apply to streaming data, such as Twitter messages or RSS feeds, providing the opportunity to index content in near-real-time, based on references to places.

However, while finding references to places in Web documents, ambiguity and uncertainty occur. Places can share a name with other places (Paris, besides being the capital of France, refers to more than sixty places around the world¹). Places are named using common language words (Park, Hope and Independence are American cities) and proper names (Washington, Houston and San Francisco). The first type of ambiguity occurs when a place name references multiple places, and it is called Geo/Geo ambiguity, or referent ambiguity. The latter ambiguity is called Geo/Non-Geo ambiguity (referent class ambiguity), which occurs when both a location and a non-location share the same name (Amitay et al., 2004). Clough et al. (2004) suggest a third type of ambiguity, named reference ambiguity, which occurs when a place is associated to many names, like New York, NYC or The Big Apple. Ambiguity makes the resolution of references to places intrinsically context-based. Although there are important work on place-based information integration and retrieval, areas such as disambiguation are still in its infancy (Vasardani et al., 2013).

An important resource to address disambiguation is the determination of the geographic scope of the document, i.e., the set of places referenced by and relevant to the contents of the document. ‘Every document has a geographical scope” (Andogah et al., 2012). Even keyword queries to search engines can have a geographic scope (Alexopoulos and Ruiz, 2012, Silva et al., 2006), since query words embed the user's intentions in the search.

References to places can be straightforward and unambiguous as geographic coordinates or not. Other sources of geographic location information can be structured (postal addresses) or unstructured (place descriptions in text). They can also be direct (place names) or indirect (references to cultural characteristics associated to places), explicit (news headers) or implicit (“9/11”). Humans are often able to recognize references to places based on such evidence, but this association does not come so easily to automated systems. Addressing this problem is one of the pressing tasks for Geographic Information Retrieval (GIR) research.

GIR extends Information Retrieval (Baeza-Yates and Ribeiro-Neto, 1999) with use of geographic locations and metadata (Jones and Purves, 2009), taking it beyond the use of keywords. GIR studies methods and techniques for the retrieval of information from unstructured or partially structured sources, including relevance ranking, based on queries that specify both theme and geographic scope (Jones and Purves, 2008, Jones and Purves, 2009). One of the most important research subjects currently in GIR involves recognizing references to places in regular text, and also in other media, such as photos and videos (Luo et al., 2011), including implicit references. The recognition of references to places in media other than text documents is beyond the scope of this paper.

Many initiatives to tackle the GIR problem of recognizing references to places in text have arisen in the recent past, usually with varied or conflicting descriptions or terminologies, targeting various applications, or using a range of reference data. The main contributions of this survey are (1) a structured and comprehensive view on contributions to the problem of determining the geographic scope of documents, (2) a review of relevant definitions and concepts, (3) a discussion of the main techniques currently used to address the Geographic Scope Resolution (GSR) problem, and (4) a proposal for a future research agenda in the area.

This paper is organized as follows. Section 2 introduces the main application areas related to the determination of the geographic scope of documents. Section 3 discusses the terminological variations found in the literature and presents them in a structured way. Section 4 focuses on the GSR problem, presenting a set of methods and a discussion of existing techniques from the literature. Finally, Section 5 shows our conclusions and indicates future research directions for the field.