by Amal Zouaq and Roger Nkambou
This paper presents an approach for building Learning Knowledge Objects (LKO) and an ontology learning component for the automatic creation of ontologies which structure the different kinds of resources needed to compose LKOs.
The ontology learning component consists of TEXCOMON (=TEXt-COnecept Maps-ONtology) an component which transforms texts into domain concept maps and ProtÃ©gÃ©.
TEXCOMON uses textual learning objects to identify keywords and key sentences. The Stanford Statistical Parser converts key sentences into typed dependency networks which are mined to find instances of lexicon-syntactic patterns defined in a manually crafted linguistic knowledge base. Other patterns are used to determine hierarchical links, instances and attributes.
The significance of concepts is determined by their number of relationships to other concepts. Thresholds determine the concepts to include in the ontology.
Finally, the authors compare their approach to ontologies created with Text-To-Onto by performing a manual evaluation with two human domain experts.