Uso de relevância semântica na melhoria da qualidade dos resultados gerados pelos métodos indutivos de aquisição de conhecimento a partir de exemplos.

Abstract

Inductive algorithms for automatic knowledge acquisition from examples generally have a purely empirical character, do not use preliminary knowledge, observing only quantitative aspects in the generation of a knowledge base. They seek to provide knowledge bases with a minimum of size and a maximum of accuracy, without any concern with the semantic quality of the results generated. In order to preserve the automatic character of the process of knowledge acquisition, with this thesis we reduce the empirical aspect of the process, proposing a kind of preliminary knowledge, in the form of semantic relevance, which is easily represented and elicited. This knowledge, either alone or in combination with other types of preliminary knowledge, such as cost and generalization, is incorporated into inductive algorithms aimed at generating better quality knowledge bases, particularly in the semantic aspect. In addition, we use semantic relevance to define an automatic process for evaluating the semantic quality of a as well as to circumvent the problem of the great computational effort presented by the Combinatorial Neural Model. In order to increase the degree of process automation as a whole, we propose in this thesis the support environment for the automatic acquisition of A4 knowledge. The environment helps in modeling the domain in examples and preliminary knowledge and in the qualitative evaluation of the results provided by generalizing algorithms.