University of Patras
Extracting knowledge from polymorphic textual data using graph-based machine learning and transformer-based deep learning techniques
Abstract
dc:descriptionThe aim of this thesis is to develop novel approaches that extract meaningful knowledge from polymorphic (unstructured and structured) textual data. To achieve this, we rely on recent advancements from the areas of Machine Learning (ML), Graph-Based Learning, Natural Language Processing (NLP) and Deep Learning (DL). This thesis comprises a series of research contributions; specifically: (i) we introduce a novel graph-based ML framework (Graph-of-Docs), which models documents in a unified graph structure, while solving several contextual limitations of traditional machine and previous graph-based learning frameworks; (ii) we study the incorporation of Graph Theory, Representation Learning and Knowledge Graphs (KGs) in the proposed framework to improve various classification tasks, while applying this framework for several real-world applications; (iii) we study the state of the art in several NLP tasks, focusing on the benefits of utilizing the Transformer architecture in them; (iv) we propose a novel keyphrase extraction approach that is highly accurate and has low computational complexity; (v) we introduce two novel Greek NLP datasets, and a series of state-of-the-art DL models aiming to facilitate future research in the understudied field of Greek NLP; (vi) we propose a unified framework, which combines advancements from KGs and Graph-based concepts with transformer- based Large Language Models (LLMs), as to improve the performance of various text generation tasks, while reducing the limitations of LLMs including hallucinations and indecisiveness. The experiments reported in this thesis validate the effectiveness of the abovementioned research findings. The code of our models and experiments is publicly available on GitHub, while our Greek NLP datasets and DL models are hosted on HuggingFace.
Degree
thesis:*- Grantor dc:publisher
- University of Patras
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Giarelis, Nikolaos
- Γιαρέλης, Νικόλαος
Subjects
dc:subject × 26- Θεωρία γράφων
- Μάθηση σε γράφους
- Μηχανική μάθηση σε γράφους
- Γράφοι γνώσης
- Μετασχηματιστής
- Επεξεργασία φυσικής γλώσσας
- Επεξεργασία φυσικής γλώσσας στα ελληνικά
- Μηχανική μάθηση
- Βαθιά μηχανική μάθηση
- Γλωσσικά μοντέλα
- Graph theory
- Graph-based learning
- Graph-based machine learning
- Knowledge graphs
- Transformer
- Natural language processing
- Greek natural language processing
- Machine learning
- Deep learning
- Language models
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Επιστήμες Μηχανικού και Τεχνολογία
- Τεχνολογία μέσων
- Electrical Engineering, Electronic Engineering, Information Engineering
- Engineering and Technology
- Media Technology
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/58627