Automatic Generation of Ontology from the Deep Web .pdf

Yoo Jung An, 1 James Geller, 2Yi-Ta Wu3and Soon Ae Chun41, 2 New Jersey Institute of Technology3 University of Michigan4 City University of New Yorkya8@njit.eduAbstractThe term “Deep Web” refers to Web pages that are notaccessible to search engines, e.g., because those Web pagesare dynamically generated in response to queries throughWeb forms or Web services. The existing automated Webcrawlers cannot index these pages, thus they are hidden fromthe Web search engines. Our goal is to properly annotatesuch Deep Web services (i.e. content generation interfaces ofhidden Web sources) with semantic indexing by constructingdomain-specific ontologies to represent the contents of theDeep Web sources. The fully automatic derivation ofontologies from Web sources without human review is to datea challenging research issue. We present a novel approach toautomatically building a large, yet domain-specific, ontologyby interweaving sub-taxonomies of WordNet with domain- specific information extracted from Deep Web service pages.Our algorithms extract domain concepts from Deep Websources which are augmented with concepts andrelationships from WordNet to construct ontology fragments.Structurally, these are Directed Acyclic Graphs (DAGs). Aniterative process of extracting WordNet concepts andrelationships and bridging concept gaps is used to tie