Uso de uma ontologia de lugar urbano para reconhecimento e extração de evidências geoespaciais na Web

• Ano de defesa: 2006
• Autor(a) principal: Karla Albuquerque de Vasconcelos Borges
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: Ontologia; Sistema de informacao geograficas; Web; Ontologias; Computação; Sistemas de informação geografica; Sistemas de recuperação da informação
• Link de acesso: http://hdl.handle.net/1843/SLBS-6XYFYS

Queries that include at least one geographic-related term, such as place names and natural features, are currently a significant subset of the queries that are submitted to search engines. Interest on local information on the Web (local search) is increasing daily, and for this kind of search, the Web is a vast repository of local geographic information. However, traditional search engines have limitation on the recognition of the geographic scope of Web pages. Pages that refer to the same place, but using alternative names, probably will not be retrieved together. Besides, in many situations the geographic context is implicit in the pages, but can be inferred by the existence, for instance, of a telephone number or postal code. In order to propose a solution for these problems, this thesis focuses on the local Web, presenting an approach based on an ontology of urban place, which allows for the recognition, extraction, and geocoding of geospatial evidences with local characteristics, such as urban addresses, postal codes, and telephone numbers as found in Web pages. The geospatial evidences are implicitly related to places, so that the contents of a page, or parts of it, can be correlated to an urban geographic location. Thus, search engines can, for instance, use such information to retrieve pages that are related to services and activities in a certain location or close to it. Therefore, the main contributions of this thesis are (1) the characterization of urban addresses contained in Web pages as sources of geospatial evidences and definition of patterns for their recognition and extraction, (2) the definition of OnLocus, an ontology of urban place that helps in the process recognizing and extracting geospatial evidences from Web pages, (3) the creation of a database for recognition of Brazilian places, based on OnLocus, (4) the proposal of a strategy for geographic categorization of a Web page, or parts of it, within a country's territorial divisions, and (5) the evaluation of the quantitative and qualitative characteristics of urban addresses that are found in the pages of the Brazilian Web. All of these contributions have been validated through experimentation, using real data from a set of 4 million Web pages. As an additional result, it was possible to obtain a snapshot of the usage of addresses in pages from the Brazilian Web and, consequently, to better understand how to geocode them. Results of this thesis open a range of perspectives for new types of applications, such as, for instance, the use of navigational links based on geographic location, geographic classification of Web pages, Web-based geospatial data mining, and semantic annotation of pages.