Learning Textologies: Networks of Linked Word Clusters

Abstract

Ontologies have been used in different important applications like information extraction, generation of grammars, query expansion for information retrieval etc. However, building comprehensive ontologies is a time consuming process. On the other hand, building a full-fledged ontology is not necessary for every application which requires modeling of semantic classes and relations between them. In this chapter we propose an alternative solution: learning a textology, that is, a graph of word clusters connected by co-occurrence relations. We used the properties of the graph for the generation of grammars and also suggest a procedure for upgrading the model into an ontology. Preliminary experiments show encouraging results.