Research On Three-way Granular Computing Approaches For Knowledge Acquisition From Dynamic Data

In the era of artificial intelligence and Big Data,for real time and efficient data mining and knowledge acquisition,the imitation of human thinking,cognition and reasoning from massive,uncertain,and complex dynamic data has become one of the hot research directions in the current field of intelligent information processing and decision-making.Granular computing is a multilevel and multiview multigranularity theory and method,which can utilize knowledge granularity under three layers of granules,granule levels and granular structures to conduct abstraction,decomposition,integration,transformation and causal analysis of the practical problem.As an emerging triarchic theory of granular computing pattern,by the simple idea of trisecting-acting and reducing complexity to simplicity,three-way decision provides an effective strategy and method of problem solving for many disciplines such as computer,medicine,management,and cognitive science,etc.Based on granular computing and three-way decision,a theoretical framework and approaches of sequential three-way decision are proposed for processing dynamic data from idea,model,and algorithm in three-way Bayesian decision-theoretic rough set model by the technologies of incremental updating and matrix operations.The specific research works are listed as follows:(1)The idea of sequential three-way decision with decision-making semantic interpretation is analyzed for dynamic decision-making data.The multilevel granular structure based on attributes sequence,the various models of selection based on seven kinds of regional combination,and the diversified cost structure based on the processing and transformation of granularity are discussed.Besides,the sequential three-way decision updating models and algorithms are designed by the technologies of incremental learning.Then a unified framework of sequential three-way granular computing and incremental processing is presented to improve the speed and reduce the risk for decision-making.It provides a practical way to simulate human's analysis,processing,and problem solving in complex and dynamic information.(2)The four levels of changes with multi-dimensional granularity are analyzed in dynamic decisionmaking information system for multi-dimensional dynamic decision-making data.The computational model of three-way decision is presented based on matrix.Besides,The knowledge matrix updating mechanism of multi-granularity hierarchical space is discussed in the dynamic evolution of object,attribute,condition attribute value and decision attribute value.The incremental updating theorems of relation matrix,intersection matrix and non-intersection matrix are constructed.The incremental updating framework of sequential three-way granular computing is proposed to improve the efficiency of dynamic three-way knowledge acquisition.It provides a novel idea and method of incremental system learning for large-scale uncertain dynamic data mining and knowledge acquisition.(3)Cost-sensitive decision problems are analyzed in the traditional two-way and three-way Bayesian models for multi-class dynamic decision-making data.Besides,a fusion matrix method with Bayesian decision from two-way to three-way is given.By considering the combinations of region for processing multi-class decision,the framework and algorithm of multi-class three-way granular computing are proposed,which expand the applicable scope of the existing theory and method in three-way decision.It provides a new kind of conflict coordination solution in costsensitive multi-class problem of three-way decision.(4)The temporal model,spatial model and both fusion of integration model are analyzed for hybrid dynamic decision-making data.A computational method of sequential three-way decision is proposed based on the fusion of temporality and spatiality.By considering the constructions of multilevel granularity space for hybrid dynamic data,a hybrid multigranularity fusion strategy is presented based on decision-making attitude and the combinations of interpretation and union operations.Moreover,a composite algorithm of sequential three-way granular computing for processing hybrid dynamic data is designed to optimize the efficiency of hybrid dynamic granularity fusion.It provides a novel dynamic multigranularity fusion approach for hybrid data decision-making.This dissertation systematically studies the decision-making knowledge acquisition of dynamic data from the perspective of granular computing and three-way decision.Not only a sequential threeway decision theoretical framework is established by effectively integrating the multi-granularity and sequential decision-making ideas into the dynamic three-way decision,but also the corresponding models and algorithms are proposed to process the multi-dimensional,multi-class and hybrid dynamic data.Besides,the effectiveness of proposed models and the efficiency of proposed algorithms are verified by experiments,respectively.This work promotes the research and development of theories and methods in three-way decision.It provides new research ideas and methods for dynamic Big Data intelligent decision-making.