多概念格集成与知识获取方法研究

Concept lattice theory is an effective method for data analysis, which has been successfully applied in knowledge discovery and other fields. However, traditional single concept lattice shows its limitation for the aquirements of more and more complex data analysis. To address this issue, we aim to develop esembling theory of concept lattices, which has important theoretical significance and applicabale value. The project is to study the analysis description framework, modeling method and knowledge acquisition of multi-concept lattice ensemble in the context of single/multiple source information systems and distributive information systems. Research content mainly includes: (1) building the connection operators in each of juxtaposed and superimposed formal contexts, and forming a description framework of data analysis of multi-concept lattice ensemble; (2) giving the judgment theorem of multi-concept lattice reduction, and designing the methods of concept selection and lattice selection in multiple concept lattices; (3) based on multi-concept lattice, researching rule acquisition method and rule redundancy detection method, proposing frequent itemsets and association rules mining algorithm, and designing efficient grade rule acqusition algorithm. This research will lay the foundation for establishing a system of knowledge acquisition based on multi-concept lattice ensemble, and provide a new theoretical foundation and technical support for solving practical problems using intelligent computing.

概念格是数据分析的一种有效方法，已被成功应用到知识发现等领域，然而面对日益复杂的数据分析需求，传统的单一概念格具有其局限性，研究多概念格集成具有重要的理论意义和应用价值。本项目拟针对单一信息系统、多源信息系统、分布式信息系统，研究多概念格集成的分析描述框架、建模方法以及知识获取，具体内容包括：（1）构建并置和叠置形式背景上的连接算子，建立多概念格集成的分析描述框架；（2）给出多概念格约简的判定定理，设计多概念格的概念选择和格选择方法；（3）给出基于多概念格集成的规则提取算法和规则冗余检测方法，提出基于多Iceberg概念格的频繁项集和关联规则挖掘算法，给出多优势概念格分级规则获取算法。本课题的研究将建立多概念格集成及知识获取的理论和方法体系，为运用智能计算解决实际问题提供新的理论基础与技术支撑。

概念格是进行数据分析的一种有效方法,已被成功应用到知识发现等领域,然而面对日益复杂的数据分析需求，传统的单一概念格具有其局限性，研究多概念格集成具有重要的理论意义和应用价值。项目组按照项目计划书的要求开展了研究工作，并取得了一系列研究成果。（1）给出了粒度概念格模型及其构造方法,证明了粒度概念格和原概念格之间的关系。给出了渐进式伪规则获取方法，提出了两个伪规则集进行合并的方法。（2）提出了一种无冗余决策规则的概念，并且给出了粒度规则和决策规则的获取方法。（3）提出了一种决策形式背景中对象压缩的理论框架和方法，保证了在不丢失决策规则信息前提下的对象约简。（4）提出了一种样本约简和维度约简的加速策略，该方法能够加速模糊粗糙特征选择的过程。构造了Shannon熵和互补熵意义下的区分矩阵，揭示了不同意义下属性约简的相互关系。（5）提出了一种结合无监督学习的数据流分类算法，通过比较分类和聚类结果的准确率，来判断是否发生了概念漂移；提出了一种基于Kappa系数的数据流分类算法。