Indexação automática por atribuição de artigos científi cos em português da área de Ciência da Informação

Abstract

This work proposes and evaluates a process of automatic indexing by assignment in the representation of full-text articles written in
Portuguese, in the context of construction of a scientifi c database in the area of Information Science in Brazil. It uses the exploratory,
bibliographic and empirical research as a methodology. The empirical part takes base in the accomplishment of an experiment as a
case study. The experiment consists of the application of the proposed process in a corpus composed of 60 scientifi c articles, as well
as quality assessment in automatic indexing through indexes of consistency, precision, recall, and F-measure. The gold standard
was the authors’ keywords. The automatic indexing process uses the Brazilian Thesaurus of Information Science and SISA software.
The satisfactory results were a consistency index average of 19%, an average precision of 30%, an average recall of 37%, and a mean
F-measure of 30%. The analysis of the results shows the thesaurus has a strong infl uence on the results of an automatic indexing by assignment, although the general term’s relations had poor contribution on the quality of the automatic indexing. In addition, we point out intervening factors in automatic indexing