Search performance can be greatly improved by describing data using Natural Language Processing (NLP) tools to create new metadata and domain ontologies. A methodology is presented to use domain specific knowledge to improve user request. This knowledge is based on concepts, extracted from the document itself, used as “semantic metadata tags” in order to annotate XML documents. We present the process followed to define and to add new XML semantic metadata into the digital library of scientific theses. Using these new metadata, an ontology is also constructed by following a methodology. Effective retrieval information is obtained by using an intelligent system based on XML semantic metadata and domain ontology.