Automatically Inducing Ontologies From Corpora

The emergence of vast quantities of on-line information has raised the importance of methods for automatic cataloguing of information in a variety of domains, including electronic commerce and bioinformatics. Ontologies can play a critical role in such cataloguing. In this paper, we describe a system that automatically induces an ontology from any large on-line text collection in a specific domain. The ontology that is induced consists of domain concepts, related by kind-of and part-of links. To achieve domain-independence, we use a combination of relatively shallow methods along with any available repositories of applicable background knowledge. We describe our evaluation experiences using these methods, and provide examples of induced structures.