Semantic Mapping of XML Tags using Inductive Machine Learning
Lukasz Kurgan, Waldemar B. Swiercz, Krzysztof J. Cios · 2002
In today's data-centric world many applications rely on data that comes from multitude of different sources. To integrate that data two major operations are performed: finding semantic mapping between data sources, and transforming structure of the data sources. One of the well-established standards for storing and sharing structured and semantically described data is XML. This paper describes system called XMapper, which is used to generate semantic mapping between two XML sources that describe instances from the same domain. The described system is novel in two ways. It uses only stand-alone XML documents (without DTD or XML schema documents) to generate the mappings. It also utilizes machine learning to improve accuracy of such mappings for difficult domains. Several experiments that use artificial and real-life domains described by XML documents are used to test the proposed system. The results show that mappings generated by the XMapper are highly accurate for both types of XML sources. The generated mappings can be used by a data integration system to automatically merge content of XML data sources to provide unified information for a data processing application.