预测模型标记语言PMML正被许多数据挖掘组织作为标准化的数据挖掘模型描述语言.然而,由于数据挖掘技术的不断发展,参与建立PMML的数据挖掘厂商的经验有差异,PMML本身含有的大量语言元素不可避免地带来基于PMML的数据挖掘元数据的语义不一致问题.为解决这个问题,提出了一种基于描述逻辑的扩展预测模型标记语言EPMML,详细分析了EPMML的描述逻辑基础SOIN,设计EPMML的语言元素.基于EPMML描述的数据挖掘元数据可以转化为基于SOIN的知识库,进而进行知识推理以自动发现数据挖掘元数据的内在语义不一致问题.Racer推理实例验证了EPMML语言的良构性,良好表达能力和推理有效性.%Predictive Model Markup Language PMML is currently used as standardized description language for data mining model by more and more DMG members. However, different experience of data mining products providers, constant development of data mining techniques, and PMML containing lots of language elements inevitably lead to inconsistency problems in PMML based data mining metadata. Considering this problem, a description logic SOIN is designed in this research. Its syntax and semantics are analyzed. An extended predictive model markup language EPMML is then proposed based on the SOIN infrastructure, the language elements are designed in detail. EPMML based data mining metadata can be transformed into knowledge base of SOIN, and then potential semantic inconsistency problems in the metadata can be automatically discovered by knowledge reasoning upon the knowledge base. Illustrations in the reasoning engineer Racer validate the well-formedness, well expressibility and reasoning efficiency of EPMML.
展开▼
