A terminologia em Sistemas de Recuperação da Informação baseada na WORDNET.PT

The article results from post-doctoral research conducted in Universidade Federal de Santa Catarina. The goal is to propose information retrieval strategies based on natural language processing, to extract semantic relations from WordNet.Pt, and use them to represent documents and users’ search expressions. The approach is qualitative, exploratory and applied to ambiguity problems in information retrieval. As for the procedures used, it is a bibliographic search. The discussion is motivated by the problem of low precision and high recall in user searches, influenced both by the absence of semantic correspondence between search expressions and terms used in indexing and by the lack of determination of the semantic similarity between document terms that, even being lexicographically different, have the same meaning. The research core is justified by the advantage of developing systems that combine natural language and controlled language, for an interactive search. Although in a partial way, the research points to important results in the solution of lexical ambiguity, through semantic relationships in the representation of documents and user search. On the one hand, this success guarantees the restriction of the search space and, consequently, precision. On the other hand, the expansion of consultations by suggesting equivalent terms from controlled vocabularies and the natural language and its variants.