Ontology Learning from definitional sentences: a knowledge graph based pipeline

Ontology Learning (OL) is the task of fully or partially automating the extraction of ontological axioms from natural language text, thus providing support for all the engineering processes underlying the construction of a domain-specific ontology. This thesis provides an overview over the research field of Ontology Engineering, giving some background on the Semantic Web technologies for ontology formalization and knowledge representation, namely RDF, OWL and SPARQL. Moreover, it describes an agile design methodology called eXtreme Design, which fosters the reuse of modular design patterns grounded in foundation ontologies like DOLCE Ultralite. The goal of the thesis is to contribute to the extension of the FrODO OL tool, by designing a pipeline which includes a classifier for definitional sentences and a new set of pattern-based heuristics for the extraction of taxonomic relations. The evaluation results of the sentence classifier are comparable to the current SOTA score for such task. The functionality of the proposed heuristics is tested on some example definitions, with the obtained results showing their feasibility for the generation of reliable ontology drafts.