Improving Dissimilarity Functions with Domain Knowledge; Applications with IKBS System

摘要

Some of the fundamental and theoretical issues in Knowledge Discovery in Database (KDD) rely on knowledge representation and the use of prior and domain knowledge to extract useful information from data. In many data exploration algorithms, dissimilarity functions do not use domain knowledge for the cases comparison. The Iterative Knowledge Base System (IKBS) has been designed to improve generalization accuracy of exploration algorithms through the use of structural properties of domain models. A general mathematical framework for utilizing structural properties of the domain model encompassing the definition of a Dissimilarity Function for Structured Descriptions is proposed. Applications are conducted with the help of IKBS on a set of databases from the UCI machine learning repository and on structured domain definition data.