Extraction de sous-ontologies autonomes par fermeture des opérateurs hyponymie et hyperonymie

Facing the exponentially increasing amount of data, new challenges consist in retrieving the good information at the right time, organizing and filtering data, visualizing them and using them in a specific decision context. For a decade, ontologies have been successfully used as semantic guides for these tasks. Nevertheless the size of ontologies that are shared and accepted as standards in a given domain may rapidly grow beyond the human capacity to grasp information. Dealing with large ontologies proved to be problematic for applications that require user interactions such as ontology-based document annotation or ontology modification/evolution. This problem can be partially overcome by providing the user with a sub-ontology focused on his/her task. Focusing on "is-a" relationships, an ontology can be represented as a direct acyclic graph. Given a reference-ontology and a set of user's concepts of interest, this paper proposes a formal definition of relevant concepts. This set of relevant concepts is made of user's concepts of interest plus some of their hyponyms and hyperonyms. Those extra concepts are added for enlighten user's concepts relationships in the resulting sub-ontology to make it self-explanatory. The set of those relevant concepts is defined using the closure of classical graph operators i.e.: least common ancestor (lca) and greatest common descendant (gcd). Efficient algorithms are also provided to identify relevant concepts and to extract the corresponding self-explanatory sub-ontology. The resulting program, called OntoFocus, may be freely used at the address http://www.ontotoolkit.mines-ales.fr/.