粗糙集理论中数据约简方法在电子商务中的应用研究

发布时间：2018-09-04 19:41

【摘要】：粗糙集理论作为一种数学工具,能处理知识的模糊性和不确定性等问题。求核属性和属性约简是粗糙集理论较为集中研究的课题,核属性是所有属性中最为核心的部分,在整个属性约简甚至最终的规则提取集中起到至关重要的作用；属性约简的目的是通过删除不相关或不重要的属性用尽量少而精的信息来表达原数据所表达的信息,已经被证明是NP-hard问题。本文在分析常用的求核属性和属性约简算法的优缺点时发现,在众多算法中大多只适用于相容决策表,而对决策表的不相容性考虑的甚少。本文提出了求核属性和属性约简的分级差别矩阵算法,根据决策表是否相容而进行不同的处理。在求核属性中,因为处理不相容决策表时,现有文献提出的改进的差别矩阵求核方法比较合理和有效,所以保留其优点,在其思想的延伸下,提出分级差别矩阵方法,新方法是通过决策属性的值进行划分,即论域的划分,通过划分的对象域形成分级差别矩阵,以分级差别矩阵和原有的差别矩阵得到的核可能是核属性为前提,确定最终的核属性。处理相容决策表时,原有方法无法得到差别矩阵时可直接用本文的分级差别矩阵求核。两差别矩阵求核方法有各自的优缺点,但是又有一定的联系,实例证明本文提出的分级差别矩阵在原有差别矩阵得不到核的情况下,可以求出属性核,证明了算法的有效性。把提出的分级差别矩阵运用到属性约简方法研究中,以求核方法中得到的可能核为出发点,求得约简集,获得决策表的约简模型。实例分析验证了两个算法的有效性。同时研究这两个算法在电子商务数据约简中的实际应用。
[Abstract]:As a mathematical tool, rough set theory can deal with the fuzziness and uncertainty of knowledge. Finding kernel attribute and attribute reduction is a research topic of rough set theory. Kernel attribute is the core part of all attributes, which plays an important role in the whole attribute reduction and even the final rule extraction set. The purpose of attribute reduction is to express the information expressed by the original data by deleting irrelevant or unimportant attributes with as little and fine information as possible. It has been proved to be a NP-hard problem. After analyzing the advantages and disadvantages of the commonly used kernel attribute and attribute reduction algorithms, it is found that most of the algorithms are only applicable to the compatible decision table, but the incompatibility of the decision table is seldom considered. In this paper, a hierarchical discernibility matrix algorithm for kernel attribute and attribute reduction is proposed, which is treated differently according to the compatibility of decision table. In the kernel attribute, because the improved discernibility matrix kernel method proposed in the existing literature is more reasonable and effective when dealing with the incompatible decision table, the advantages of the improved discernibility matrix method are preserved, and the hierarchical difference matrix method is put forward under the extension of its thought. The new method is to divide the decision attribute by the value of decision attribute, that is, the division of the domain, and form the hierarchical discernibility matrix by dividing the object field. The kernel obtained by the hierarchical discernibility matrix and the original discriminant matrix may be the kernel attribute. Determine the final kernel attribute. When dealing with compatible decision table, the kernel can be directly obtained by using the hierarchical discriminant matrix when the original method can not get the discernibility matrix. The two difference matrix kernel method has its own advantages and disadvantages, but also has certain relations. The example proves that the hierarchical difference matrix proposed in this paper can work out the attribute kernel when the original difference matrix is not kernel, and proves the validity of the algorithm. The hierarchical difference matrix is applied to the study of attribute reduction. The reduction set is obtained and the reduction model of the decision table is obtained based on the possible kernels obtained from the kernel method. The effectiveness of the two algorithms is verified by an example. At the same time, the practical application of these two algorithms in e-commerce data reduction is studied.
【学位授予单位】：东北林业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F724.6;TP18

【参考文献】