Content and data linking leveraging ontological knowledge in data journalism. (Utilisation de connaissances ontologiques dans la liaison de contenus et de données appliquée au journalisme de données)

Cheikh Brahim El Vaigh

Abstract: This thesis is about the creation of links between textual content and ontological knowledge bases (KBs). It pertains several areas of research: natural language processing, information retrieval and semantic web, and in particular RDF-based KBs. We propose to study collective entity linking, which consists in linking at once mentions of entities present in a textual document to entities in a KB. To that end, we leverage semantic measures, i.e., entity relatedness measures which exploit the relationships between the entities in a KB. We contribute by the definition of well-founded entity relatedness measures that benefit to the extent possible from the properties of RDF KBs through (basic) reasoning, and thus allow to improve the state-of-the-art. Furthermore, we are also interested in the alignment of different KBs, based on KBs embedding techniques. This alignment not only allows to enrich the KBs at hand, but also to indirectly improve the collective entity linking. We contribute by an alignment criterion, based on the alignment of the dimensions of the KBs embedding spaces, which, notably does not need any prior knowledge to perform said KBs alignment.