Modelo para recupera??o de informa??o em reposit?rios institucionais utilizando a t?cnica de sumariza??o a partir da sele??o de atributos do Cassiopeia

Submitted by Jos? Henrique Henrique (jose.neves@ufvjm.edu.br) on 2020-04-07T19:41:45Z No. of bitstreams: 2 license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) luanna_azevedo_cruz.pdf: 1783931 bytes, checksum: 076cc0b726975db741a999816b3bc05f (MD5)Approved for entry into archive by Jos? Henrique Henrique (jose.neves@ufvjm.edu.br) on 2020-04-07T19:42:22Z (GMT) No. of bitstreams: 2 license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) luanna_azevedo_cruz.pdf: 1783931 bytes, checksum: 076cc0b726975db741a999816b3bc05f (MD5)Made available in DSpace on 2020-04-07T19:42:22Z (GMT). No. of bitstreams: 2 license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) luanna_azevedo_cruz.pdf: 1783931 bytes, checksum: 076cc0b726975db741a999816b3bc05f (MD5) Previous issue date: 2019Um grande reposit?rio de cultura e conhecimento vem sendo formado a partir de documentos digitais criados por usu?rios das mais diversas ?reas. No contexto educacional, textos acad?micos comp?em bases textuais, como os reposit?rios institucionais, que s?o fonte de informa??o e auxiliam no processo de ensino e aprendizagem. No entanto, assimilar e lidar com o elevado volume de informa??o dispon?vel, localizando-as de forma r?pida e precisa, passou a ser um desafio. Neste sentido, a ?rea de Recupera??o de Informa??o atua com o prop?sito de detectar, dentre uma cole??o de documentos, os que satisfazem ?s necessidades do usu?rio. Por?m, problemas como quantidade excessiva de documentos retornados e falta de relev?ncia e precis?o dos resultados apresentados dificultam o processo de recupera??o de informa??o. T?cnicas de Minera??o de Textos podem auxiliar nesse processo, por meio da extra??o de dados, descoberta de padr?es, associa??es e regras, realiza??o de resumo, e an?lises em documentos de texto. Dessa forma, o objetivo desta pesquisa ? analisar se a aplica??o da t?cnica de sumariza??o, a partir do m?todo de sele??o de atributos (palavras) do modelo Cassiopeia (implementado no sumarizador PragmaSUM), num corpus de textos acad?micos, auxilia na recupera??o de informa??o, diminuindo a sobrecarga de informa??o e melhorando a precis?o dos resultados retornados ao usu?rio. A sele??o de atributos do modelo consiste em um m?todo de redu??o da alta dimensionalidade e dados esparsos. A pesquisa foi desenvolvida em seis etapas que compreenderam as seguintes a??es: levantamento bibliogr?fico; prepara??o do corpus e sumariza??o dos textos acad?micos; implementa??o de um buscador; execu??o da recupera??o de informa??o padr?o e com a sele??o de atributos do modelo Cassiopeia; avalia??o da recupera??o de informa??o por interm?dio das m?tricas precision, recall, e F-measure; e, por fim, an?lise dos dados a partir dos testes estat?sticos ANOVA de Friedman e coeficiente de concord?ncia de Kendall. Os resultados obtidos mostraram que a sumariza??o, efetuada principalmente com altas taxas de compress?o (80% e 90%), diminuiu a sobrecarga de informa??o e aumentou a precis?o dos resultados apresentados ao usu?rio, permitindo qualidade na recupera??o de informa??o em textos acad?micos. Al?m disso, simplificou o processo de indexa??o, atenuou a alta dimensionalidade e promoveu maior agilidade na recupera??o de informa??o.Disserta??o (Mestrado Profissional) ? Programa de P?s-Gradua??o em Educa??o, Universidade Federal dos Vales do Jequitinhonha e Mucuri, 2019.A great repository of culture and knowledge has been formed from digital documents created by users from various ?reas. In the educational context, academic texts make up textual bases, such as institutional repositories, which are a source of information and assist in the teaching and learning process. However, assimilating and dealing with the high volume of information available, locating it quickly and accurately, has become a challenge. In this sense, the Information Retrieval area acts with the purpose of detecting, among a collection of documents, those that satisfy the user's needs. However, problems such as excessive amount of returned documents and lack of relevance and accuracy of the results presented make the information retrieval process difficult. Text Mining Techniques can assist in this process by extracting data, discovering patterns, associations and rules, summarizing, and analyzing text documents. Thus, the objective of this research is to analyze if the application of the summarization technique, based on the attributes selection method (words) of the Cassiopeia model (implemented in the PragmaSUM summarizer), in a corpus of academic texts, helps in the retrieval of information, reducing information overload and improving the accuracy of results returned to the user. Model attribute selection consists of a method of reducing hogh dimensionality and sparse data. The research was developed in six steps that comprised the following actions: bibliographic survey; corpus preparation and summarization of academic texts; implementation of a search engine; performing standard information retrieval and selecting attributes from the Cassiopeia model; assessment of information retrieval using precision, recall, and F-measure metrics; and, finally, data analysis from Friedman ANOVA and Kendall Agreement coefficient statistical tests. The results showed that summarization, mainly performed with high compression rates (80% and 90%), reduced the information overload and increased the accuracy of the results presented to the user, allowing quality information retrieval in academic texts. In addition, it simplified the indexing process, attenuated the high dimensionality and promoted faster information retrieval