Abstract:

In this paper, a novel computational approach for knowledge granulation and its meaning are investigated from a new point of view of matrix. Firstly, two computational approaches for knowledge granulation, discernibility degree and attribute importance are proposed based on the equivalent relation matrix. Secondly, the relationship between the equivalent relation matrix and the knowledge granulation is analyzed, and hence the relevance between the equivalent relation matrix and the uncertainty of an information system is revealed. Furthermore, the meaning and the inherent essence of the matrix expression of knowledge granulation are analyzed. Finally, the matrixbased computational method for attribute importance is applied to calculate the core and the minimum reduction of attribute combined with the update of equivalent relation matrix while one attribute added to or removed from the attribute set, the numerical examples demonstrates the effectiveness of matrixbased computational method of attribute importance.  

Key words: knowledge granulation;discernibility degree;attribute importance;equivalent relation matrix;reduction