动态多标记决策信息系统下基于序贯三支决策的最优标记选择

发布时间：2018-11-08 12:48

【摘要】：实际应用中,随着各种数据观测工具、实验设备效率与性能的提高,以及对于数据的搜集方式与处理方法的多样性增长,我们获取到的数据普遍呈现持续增长、不断更新、不断细致化的动态现象。数据的不断更新与细致化,将会导致原有信息粒、知识结构的动态变化,时效性成为动态数据环境中知识获取的关键问题。Wu对于对象在不同尺度下拥有不同的知识粒度,提出了多标记信息系统。仔细观察不难发现,这样的多标记信息系统其实就是一个信息不断更新变化的过程。然而在现实生活中,虽然足够详细的数据信息能够使我们更加清晰地认识事物,但在某种程度下,搜集处理这些信息的代价也是很大的,甚至庞大的信息中存在的无用价值信息也会过多地耗费成本,这种情况下再去不必要地细致化信息是得不偿失的,因而我们把序贯三支决策引入到多标记信息系统,在多标记决策信息系统下研究了通过动态决策来选取最优标记的相关问题。最优标记选择是多标记决策信息系统中的一个重要问题,然而由于现实生活中我们所研究的多标记决策信息系统往往会出现对象更新和属性更新的情形,因而现有的最优标记选择方法并不总是适用的。同时我们也发现,序贯三支决策是研究信息更新的一个重要手段,因此在本文中,我们用序贯三支决策方法研究了动态多标记决策信息系统下的最优标记选择问题。详细来说,首先我们将序贯三支决策引入多标记信息系统,这样的多标记信息系统可以看作是论域的多粒度表示。然后,加入决策属性就有了多标记决策信息系统,类似的引入序贯三支决策。最后,考虑到多标记决策信息系统下的数据更新,我们研究了数据更新下不确定域的变化情形,进而分析了最优尺度的变化问题。同时,我们也在最后给出了一些数据实验来验证本文最优尺度选择的有效性。
[Abstract]:In practical applications, with the improvement of the efficiency and performance of various data observation tools, experimental equipment, and the diversity of data collection and processing methods, the data we have obtained have generally been continuously growing and updating. The dynamic phenomenon of constant refinement. The continuous updating and meticulous of data will lead to the dynamic change of original information grain and knowledge structure, and timeliness becomes the key problem of knowledge acquisition in dynamic data environment. Wu has different granularity of knowledge for objects at different scales. A multi-label information system is proposed. It is not difficult to observe that such a multi-label information system is a process of information updating and changing. However, in real life, while sufficiently detailed data can enable us to understand things more clearly, to some extent, the cost of collecting and processing this information is also enormous. Even the useless value information that exists in the huge information will consume too much cost. In this case, it is not worth the gain to turn the information into detail unnecessarily, so we introduce three sequential decisions into the multi-label information system. In this paper, the problem of selecting the optimal label by dynamic decision is studied in the multi-label decision information system. Optimal label selection is an important problem in multi-label decision information systems. However, the multi-label decision information systems we study in real life often have object update and attribute update. Therefore, the existing optimal tag selection methods are not always applicable. At the same time, we also find that sequential three-branch decision making is an important means to study information updating. In this paper, we use sequential three-branch decision method to study the optimal label selection problem in dynamic multi-label decision information system. In detail, first of all, we introduce the sequential three-branch decision into the multi-label information system, which can be regarded as the multi-granularity representation of the domain. Then, there is a multi-label decision information system by adding decision attributes, which is similar to the sequential three-branch decision-making. Finally, considering the data update in multi-label decision information system, we study the change of uncertain domain under data update, and then analyze the change of optimal scale. At the same time, we also give some data experiments to verify the effectiveness of the optimal scale selection.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：N945.25

【参考文献】