面向领域用户知识发现的数据结构化建模与多粒度融合

In the age of big data, that how to intelligently and efficiently discover domain user's knowledge has been becoming a key problem in current research, and multi-granularity knowledge discovery model based on data-driven has gradually become the new key ideas in the data science. Big data always is huge, heterogeneous, sparse, high dimensional, dynamic and real-time. Aiming at the above features of big data, this project will propose a multi-granularity data mining model for processing big data based on both data feature, user's requirements and the multi-granularity cognitive mechanism of human being. The main contents of this research project include: Based on cloud model and support vector machine(SVM), a multi-granularity automatic preliminary classification model will be established for multi-source heterogeneous data; According to the human brain mechanism of comprehensive expression and processing knowledge in multi-granularity levels, a multi-granularity relationship virtual structured representation model of big data will be constructed; Based on human brain information fusion mechanism in different granularity levels for different information sources, a progressive multi-granularity data fusion model will be proposed; According to domain-oriented user's requirements, this project will realize personalized knowledge acquisition from the fusion data by rough set theory, and research the evolution rules of both knowledge's denotation and its connotation with the dynamic change attributes and objects.This research project may help to establish a multi-granularity framework of big data deep-learning and knowledge discovery. Furthermore, it may also promote the application of granular computing theory under big data environment, and improve the research of both computing science and data science.

大数据时代，领域用户如何智能高效地获取知识成为当前数据科学面临的难题，基于数据驱动的多粒度知识发现模式逐渐成为数据挖掘的新思路。针对复杂大数据具有异构、稀疏、高维、动态等特征，本课题利用数据特征和用户需求联合驱动的机理，结合人脑求解复杂问题的多粒度模式，研究大数据环境下的多粒度知识发现理论模型和方法。主要研究内容有：面向领域用户需求，融合云模型、支持向量机等方法，建立复杂大数据的多粒度自动初分类模型；针对初分类数据，模拟人脑从多个层次进行认知的综合表达与处理机制，建立多粒度关联式虚拟结构化数据表示模型；根据人脑能够在不同粒度层次上融合不同来源信息的机理，构建多粒度递进式数据融合模型；利用粗糙集等方法提取领域用户个性化知识，并研究知识内涵和外延的演化规律。本课题的研究将有助于建立大数据深度学习与知识发现的多粒度模型框架，促进粒计算理论在复杂大数据环境下的应用，推进计算科学和数据科学的发展。

动态数据环境下，面向领域用户需求的多粒度知识发现模式逐渐成为智能数据分析的新思路。本课题利用数据特征和用户需求联合驱动的机理，结合多粒度认知机理，研究复杂数据环境下的多粒度知识发现理论模型和方法。主要在数据的多粒度分类与信息约简、不确定数据的近似描述与多粒度结构化建模分析、数据的多粒度信息融合与最优化信息处理、面向不确定性数据的高效知识发现等方面进行深入研究，并取得了一系列原创性研究成果。发表相关学术论文60篇，其中SCI收录的期刊论文42篇，EI收录的期刊论文9篇，会议论文3篇（1篇被SCI检索），中文核心期刊收录6篇；出版学术著作3部；申请发明专利8项。本课题的研究有助于建立复杂大数据环境下的深度学习与知识发现多粒度模型框架；设计并实现的多粒度高效知识发现算法，有助于推进计算科学和数据科学的发展。