Geographically-Aware Information Retrieval on the Web

Introduction

Numerous contributions have been made in the field of Information Retrieval (IR) since the 60s to maximise the amount of relevant information which can be retrieved from large collections of information resources. However, with the rapid growth of the World Wide Web (WWW) since the 90s, additional challenges have been posed to the IR task. Moreover, the Web users have been changing gradually the way they use the Web. People are now connected to each other via social networks and are using mobile devices intensively. It has led to considerable modification in the way the users look for information and evaluate their relevance. This new scenario has motivated the development of specialised Web search engines which aim to meet particular user needs.

Extremely roughly, traditional Web search engine algorithms are based on keyword matching, which imposes a number of limitations to perform more specialised IR tasks, such as the retrieval of geographic information. For example, a Web document containing the sentence “...with the establishment of the company in Leeds, thousands of medical equipments will be manufactured daily...” would not be retrieved by a typical search engine by querying the system using the keywords “medical equipment manufacturer england,” since no word in the document’s text would match the term ‘england’ (unless, of course, some other text fragment in the document also contains the word ‘England’). This happens because the expression ‘england’ is treated as an ordinary term, not as a geographic place. In a geographically-aware search engine this document should be returned, since the system should be able to infer that the term ‘Leeds’ refers to a city which is located in a country which can be referred to by the term ‘england’.

Geographic Information Retrieval (GIR) is a recent research area which has become notably attractive. Geographic Web search engines are specialisations of standard Web search engines, adding to them the ability to identify geographic contexts of Web resources (e.g., texts, images, movies) and to index them according to such contexts. This article presents the main topics of research within GIR and overviews significant contributions in the field.

Background

Most of the information available on the Web has some sort of geographic context. This context includes the place where the information has been created, places referred to in the document, and those in which the information is regarded as more relevant. Some experiments reported in the literature have demonstrated that a considerable amount of Web pages contains terms which may be derived into geographic places. Examples of such terms include place names, telephone numbers and postcodes.

McCurley (2001) observed that approximately 8.5% of Web pages contain a telephone number, 4.5% contain a postcode, and 9.5% contain one of the two. Silva et al. (2006) analysed 3,775,611 Web pages and noticed that they contain an average of 2.2% references to Portuguese cities (obviously, if other place names apart from cities had been taken into account, such numbers are likely to be even greater). Nevertheless, traditional information retrieval systems do not consider this context in their parsing, indexing and retrieval process. This section presents the main motivations for the development of GIR. For an appropriate understanding of the field, it is categorised here into different sub-areas and the major challenges and problems faced in each of them are described.