Multigranulation-based Knowledge Updating Approaches In Dynamic Data Environments

In the era of internet technology,there are more and more complex data collected from various research areas.The data are diverse and dynamic.How to efficiently and reasonably cope with these dynamic complex data for data mining and knowledge discovery by making use of uncertainty reasoning techniques,and further obtain the potential valuable knowledge is becoming one of the hot research directions in the area of intelligent information processing.The theory of multigranulation rough sets makes use of the information granules and multiple granular structures obtained from information systems to characterize a target concept for complex data analysis,and further mine potentially useful knowledge.This theory provides an effective theoretical basis for complex data analysis and intelligent decision making.The data in real application change over time,such as financial analysis,student performance evaluation.The existing dynamic updating algorithms mainly update knowledge from the perspective of single granular structure.These algorithms are not suitable for updating knowledge in multigranulation environments.How to take advantage of potential obtained information of the multiple granular structures to obtain multigranulation knowledge,and further improve the efficiency for updating multigranulation decision knowledge,is one of the key challenge.Based on the theory of multigranulation knowledge discovery,this dissertation focuses on exploring the matrix-based dynamic algorithms for updating multigranulation knowledge by making use of the techniques of incremental updating and matrix operations under the frameworks of neighborhood multigranulation rough sets model and dominance-based multigranulation rough sets model.The specific research works and contributions are outlined as follows:1.By considering the addition or deletion the granular structures in neighborhood information systems,the dynamic updating problems for the three-way regions in neighborhood multigranulation rough sets are investigated.In neighborhood information systems,the matrix-based representation of three-way regions in the neighborhood multigranulation rough sets is characterized by using neighborhood relation matrix.Moreover,the updating mechanisms of characteristic functions and relation matrices for neighborhood multigranulation three-way regions are proposed when adding or deleting granular structures.Based on the proposed mechanisms,the matrix-based dynamic algorithms for updating the three-way regions are developed when adding or deleting granular structures in neighborhood information systems.Compared with the matrix-based non-incremental algorithms,numerical experiments on publicly available datasets are conducted to assess the efficiency and feasibility of the dynamic updating algorithms.2.By taking into account the variation of the objects in neighborhood information systems,the dynamic updating problems for the three-way regions in neighborhood multigranulation rough sets are investigated.According to the matrix-based representation of neighborhood multigranulation three-way regions,the dynamic mechanisms of characteristic function of the target concept and the neighborhood relation matrix of each granular structure are investigated are analyzed under the variation of multiple objects.Furthermore,the updating mechanisms of intermediate matrices in each granular structure are designed.In addition,the matrix-based updating algorithms for acquiring neighborhood multigranulation three-way regions are proposed in neighborhood information systems with the variation of multiple objects.Experimental evaluations are carried out on publicly available datasets to show the efficiency and effectiveness of the dynamic updating algorithms3.By considering the variation of objects in ordered information systems,the dynamic updating problems of approximations of dominance-based multigranulation rough sets are investigated.In ordered information systems,the matrix-based representation of dominance-based multigranulation lower and upper approximations is characterized.Moreover,the incremental updating mechanisms for characteristic function of a target concept and the dominance-based relation matrix of each granular structure are proposed with time-evolving ordered data.Additionally,the dynamic updating mechanisms of the lower vector and the upper vector which are used for characterizing the dominance-based multigranulation approximations are developed.Based on the updating mechanisms,the matrix-based dynamic algorithms for updating dominancebased multigranulation approximations are proposed while the objects are added into or deleted from ordered information systems.Experimental evaluations on publicly available datasets demonstrate the effectiveness of dynamic updating algorithms.4.By considering incremental granular structures in ordered decision information systems,the dynamic updating problems of approximations in dominance-based multigranulation rough sets are investigated.Based on the decision classes in ordered decision information systems,the matrix representation of dominance-based multigranulation approximations of each decision class is characterized.The matrix-based incremental mechanisms for characteristic matrices of lower and upper approximations of each decision class are investigated when adding granular structures.Moreover,when adding an attribute set into granular structures,the updating mechanisms for dominance relation matrix are proposed.Additionally,the updating mechanisms of the lower and upper matrices of each granular structures are developed according to the variation of dominance relation matrix.Based on the updating mechanisms,when adding granular structures or adding an attribute set into each granular structure,the dynamic algorithms for updating dominance-based multigranulation approximations of each decision class are proposed.Numerical experiments on publicly available datasets show the superior performance of the proposed algorithms in terms of computational efficiency.This dissertation systematically studies the matrix-based algorithms for updating multigranulation knowledge from the perspective of theory for numerical data and ordered data with the variation of objects or granular structures.The dynamic mechanisms for updating knowledge provide effective theory and methods for uncertain data processing and analysis in dynamic data environments.