Ontology learning by using text clustering techniques: Method for structuring taxonomies

Finding an appropriate structure that represents the information contained in texts is not a trivial task. There are different structures for modeling the knowledge, such as: ontologies, taxonomies, thesaurus, and semantic networks. Ontologies are especially useful because they support the exchange and sharing of information. An important task in ontology learning is to obtain a set of representative terms to model a domain and organize them as taxonomy. For getting the representative terms on a specific domain, the linguistic analysis on text is essential. The identification of hypernymy/hyponymy relations between terms is imperative for building a taxonomy. This document introduces a novel mechanism to obtain representative terms and to find hypernymy relations between them within a knowledge domain. The basis of this idea is the use of a triple term structure <subject> verb <object> as representation model. This approach uses a set of linguistic patterns to get representative terms and combines WordNet synsets and context information for building a query set. This query set is sent to a web search engine in order to retrieve the most representative hypernym for each term. An implementation of the approach has been applied to two types of content inputs, a document corpus and a set of short texts (tweets), both of them showing promising results. This report of activities corresponds to the second year of doctoral studies.