This paper presents the three methodologies for data cleaning and semantic relationship refinement to solve the problem of producing well-defined semantics from poorly defined or underspecified semantics in a thesaurus. The system refines the semantic relationships though noun phrase analysis, WordNet alignment, and semantic relationship rules, some generated by experts and others generated from annotated examples by an inductive statistical machine learning system. Finally, the relationships were verified by the expert. Initial results are promising.
Ontologies with precise semantic are important for improving retrieval systems, for automating processes through machine reasoning, and for the Semantic Web. Developing ontologies is labor-intensive and time-consuming. This paper contributes to solving the ontology development bottleneck by exploiting the enormous intellectual capital amassed over many years in classification schemes and thesauri.
