基于粒计算的动态三支决策理论与方法

With the fast development of big data, one of the important research hotspots and directions to deal with big data is how to obtain knowledge from complex systems efficiently and in real-time. As an emerging methodology to solve decision making with uncertainty, three-way decisions utilize the ideas of triarchic methods, and provide a new approach to handle complex decision problems. This project investigates the theories, methodologies, and applications of dynamic three-way decisions, where theory of granular computing is used as the basis, decision-theoretic rough set model (DTRS) is used as objects, incremental learning is used as the means, and cloud computing framework is used as the platform. The main tasks of the project include: 1). Two types of three-way decision models are constructed when considering the changes of conditional probability and cost function in DTRS; 2). The linkage change mechanism between information table and decision risk, the incremental learning rules among three decision regions, and the selection mechanism of optimal granular layer are carefully investigated under the variation of different granules; 3). Under the variations of multiple granules, designing the parallel algorithm for the dynamic three-way decisions based on the framework of cloud computing; 4). Discussing the applications of dynamic three-way decisions in recommend systems. The project uses three-way decisions and granular computing to study the dynamic decision problems, which will enrich the theory and methodology of three-way decisions, and will further contribute to the development of decision analysis based on granular computing.

在大数据时代到来之际，如何快速实时地从复杂系统中获取决策知识，已成为大数据的重要研究热点和方向之一。三支决策作为一种新兴的处理不确定性问题的重要方法，利用三分而治、化繁为简的朴素思想，为人们处理复杂决策问题提供了一种新的途径。本项目以决策粗糙集为对象，以粒计算思想为指导，以增量学习技术和云计算并行框架为手段，来研究决策环境和粒度变化下的动态三支决策理论、方法和应用。具体研究内容为：1.构建决策粗糙集条件概率和损失函数变化下，两种动态决策粗糙集三支决策模型。2.研究不同粒度变化下信息表与决策风险的联动变化机制、不同区域决策规则增量学习算法、以及最优粒层的选择机制。3.设计多维粒度动态演化时，基于云平台的动态三支决策的并行算法。4.讨论动态三支决策在推荐系统中的应用研究。本项目从三支决策和粒计算的视角来研究动态决策问题，不仅丰富了三支决策的理论和方法，而且可推动粒计算在决策分析的应用和发展。

在大数据时代到来之际，如何快速实时地从复杂系统中获取决策知识，已成为大数据研究的重要热点和方向之一。作为一种不确定人工智能的重要方法，三支决策利用三分而治、化繁为简的朴素思想，为人们处理复杂决策问题提供了一种新的途径。本项目以决策粗糙集为对象，以粒计算思想为指导，以增量学习技术和云计算并行框架为手段，来研究决策环境和粒度变化下的动态三支决策理论、方法和应用。首先，构建了决策粗糙集条件概率和损失函数变化下，两种动态决策粗糙集三支决策模型。其次，研究了不同粒度变化下信息表与决策风险的联动变化机制、不同区域决策规则增量学习算法、以及最优粒层的选择机制。再者，设计了多维粒度动态演化时，基于云平台的动态三支决策的并行算法。最后，讨论了动态三支决策在推荐系统中的应用研究。本项目从三支决策和粒计算的视角来研究动态决策问题，不仅丰富了三支决策的理论和方法，而且可推动粒计算在决策分析的应用和发展。截止2022年12月底，项目组已发表与研究内容密切相关的学术论文40余篇，出版专著1部，获得省部级社会科学优秀成果二等奖1项，圆满地完成了项目的预期研究目标。