On the Use of Correspondence Analysis to Learn Seed Ontologies from Text

In the present work we show our approach to generate hierarchies of concepts in the form of ontologies starting from free text. This approach relies on the statistical model of Correspondence Analysis to analyze term occurrences in text, identify the main concepts it refers to, and retrieve semantic relationships between them. We present a tool which is able to apply different methods for the generation of ontologies from text, namely hierarchy generation from hierarchical clustering representation, search for Hearst Patterns on the Web, and bootstrapping. Our evaluation shows that the precision in the generation of hierarchies of the tool is attested to be around 60% for the best automatic approach and around 90% for the best human-assisted