A New Method for Knowledge Discovery of Complex Data Based on Structural Partial-Ordered Theory

Abstract

In order to develop a new knowledge discovery method with higher generalization ability, this paper proposes the generalized model of partial-ordered structure diagrams. This model is based on two philosophical methodologies: the concept driven methodology and the data driven methodology. In essence, the concept driven methodology is top-down principle, that is, the attributes representing object universality are put at the top of the structural partial-ordered diagram. While the data driven methodology is bottom-up principle in which the attributes representing object specificity are put at the top of the diagram. The method is described by the mathematical partial order theory and formal concept analysis theory. Finally, three concrete data sets are used as examples to generate diagrams. The generated diagrams can clearly reveal the knowledge implied in the complex data. It is proven that the proposed generation theory and the model constructed with partial-ordered diagrams have a good ability of generalization, and they are original methods which can be used in different domains for knowledge discovery.