Uma abordagem para a indexação semântica de documentos textuais baseada em fontes heterogêneas de informação.

Abstract

Nowadays, one of the main challenges in the area of Information Retrieval (IR) is the development of systems that correctly process the idea or concept in the queries emitted by users. Conventional IR systems usually limit their functionality to indexing and retrieving keywords, which creates incipient results when indexed terms are not mentioned in the query. Queries such as: “The king of Brazilian music” and “Roberto Carlos”, even using a distinguished group of words, may represent the same idea or concept; therefore, the system should return the same set of answers. However, for IR systems that do not consider the semantic aspect, both queries return different answering sets. In this work, we proposed a new paradigm of semantic indexing of concepts. With this new approach, concepts present in textual documents are semantic enriched automatically using information which is presented in heterogeneous sources joined in a single environment features of dictionaries, encyclopedias and common sense. In this way, the idea of object contextualization is isolated from the several forms of object representations. The proposed approach was compared with UBY project, a large scale lexic-semantic resource which combines a wide range of information built by experts and collectively for English and German languages. Both approaches were subjected to various collections of documents and was proven the superiority of the proposed approach compared to UBY. To make this conclusion we measured: the number of concepts found in the collections of documents identified by either approach; connectivity, which was computed for each element identified the number of connections established with other concepts; and quality of produced semantic enrichment, which was computed if the semantic relations between concepts established are consistent.