My Dissertation Resources
Background
Titile
Enchancing LDA for Ontology Learning
Info
- Supervisors: Mounira Harzallah and Fabrice Guillet
- University: University of Nantes, France
- Laboratory: DUKe team of LS2N (Laboratoire des Sciences du Numérique de Nantes)
- Doctoral School: MATHSTIC (École doctorale Mathématiques et sciences et technologies de l’information et de la communication (Rennes))
- Period: 2017/10 - 2021/6
- Fundings: from University of Nantes, France
Abstract
This dissertation aims to enhance LDA’s utilities of conceptualizing terms towards ontology learning, where similar terms are clustered to the predefined core concepts. We explored the classic workflow of term clustering and studied the clustering impacts of the terms representation techniques. Comparatively, we proposed the LDA based clustering strategy, where the prior knowledge embedding techniques are applied to semisupervise the LDA for the more satisfying clusters. In addition, we built up the taxonomic structure of the ontology, by internally applying the subcategorization frames over noun phrases and externally benefitting from the knowledge bases. The experiment results showed that our proposed LDA based clustering strategy outperformed the majority of the clustering works in the classic workflow. Our optimal prior knowledge embedding approach exceeded the performance of basic LDA and Seeded LDA but dropped behind the Z-label LDA. This dissertation suggests that the LDA based clustering strategy could contribute to the anticipating term conceptualizations for ontology learning.
Open Resources
Dissertation
- My dissertation official webpage.
- To download the pdf (193 pages in total), please click this link.
Final Defense
- Live Stream Video (4 or 5 hours)
- Presentations Slide
- pdf format
- latex format
- I appology that I cannot provide this source latex code without permission from my original school. However, I would like to share the Latex Slide Template that I used, which is open sourced in the Overleaf community.