一种基于频繁k元一阶元规则的多维离散数据挖掘模型

Research on Frequent k-ary Meta Rule in First Order for Multi-dimensional Discrete Data Mining

为实现对多维离散数据的挖掘,提出了包含"与"、"或"、"非"逻辑的元规则概念模型,定义了元规则实例及相应的支持度和置信度概念。在此基础上提出了新的更精炼且更有启发意义的k元一阶元规则概念模型,定义了频繁度概念,证明了k元一阶元规则的空间性质定理包括上下界计算公式。文中的元规则具有更高的抽象层次,更小的解空间,能够描述元数据间的关系以及强规则实例的分布的情况。给出了k<5时,k元一阶元规则的空间分布情况的实验结果,验证了空间性质定理。实验结果表明,在标准数据集上显著k元一阶元规则的数量比相应的强的元规则实例数少1个数量级,频繁度为100%的k元一阶元规则比强的元规则实例数少2个数量级。

To process multi-dimensional discrete data, formal concept of meta-rule including connective ＂AND＂ ＂OR＂ or ＂NOT＂ was proposed, Support degree and confidence degree of meta-rule instance were defined. Solution space of meta-rule problem was analyzed. Furthermore, formal concept of frequent k-ary Meta Rule in First Order （k-MR） was introduced. The concept of frequent degree and the hound equation of solution space of k-MR were presented, The k-MR, with smaller solution space, is more abstract than its base rule. It can represent distribution of strong meta-rule instance and relationship between meta-data. Space distribution of k-MR was also studied and verified in experimental evaluation where k 〈 5. Experimental results showed that the new method for multi-dimensional dicrete data mining was effective. On real data sets, number of meta-rule about strong meta-rule instance is about 10 times less than that of strong meta-rule instance, and number of meta-rule whose frequent degree equals 100% is about 100 times less than that of strong meta-rule instance.