Multi-level Knowledge Discovery Method Based On Formal Concept Analysis

The formal concept analysis theory was proposed by the German mathematician Wille in 1982.It starts from the formal contexts and studies the conceptual representation and knowledge discovery between data.With the deepening of research,people have introduced multi-granularity labeled methods into rough set theory and formal concept analysis,which makes the methods of knowledge discovery more diversified.At present,the research on multi-level knowledge discovery based on rough set theory is relatively mature,but the research on multi-level knowledge discovery based on formal concept analysis theory is not perfect,so this paper mainly studies multi-level knowledge discovery under the framework of formal concept analysis.The main research contents and innovations are as follows:(1)Referring to the granular labeled reconstruction method of multi-granularity labeled information system in rough set theory,we reorganize the single-granularity labeled value in the formal contexts,give the concept of the meso-granularity labled formal contexts,and prove the data structure after reorganization is the lattice structure of the original data,which greatly enriches the research topics of data mining and makes the results more diverse.(2)Due to the large number of meso-granularity labled formal contexts after reorganization,it is necessary to sort out the data structure.Therefore,the generalization and specialization of the meso-granularity labled formal contexts are proposed to reveal that the data structure can form a complete lattice.(3)This paper introduces the knowledge discovery of the meso-granularity labled formal contexts with decision-making information,and gives the single-granularity labeled method,the meso-granularity labeled method of decision implication mining and two inference evolution methods,proving two different thickness.The decision implication in a meso-granularity labled formal context can be transformed into each other,and numerical experiments have been used to verify that the meso-granularity labeled method really helps to improve the computational efficiency.