Technical University of Crete (TUC)
Γνωσιακές μηχανές με χρήση μεθόδων μηχανικής μάθησης
Abstract
dc:descriptionModern society is characterized by an unprecedented growth in the ways data and information are being produced and shared, as a result of the rapid increase in computing power, of the availability of resources and of the ability to process huge data volumes, mainly derived from Internet sources. The occurring data flood, commonly encountered in the form of natural language, inevitably reduces the recipients' collective attention span, leading more to the stressful and superficial consumption of information, rather than to its actual assimilation and evaluation. Many research groups worldwide are responding to the growing demand for automated management, representation and extraction of valuable knowledge from the continuous data streams that are overwhelming the Web, by exploiting natural language processing methodologies and tools. However, most of today’s research is disproportionally focused on around 20 of the world’s more than 7000 spoken languages, leaving the vast majority of them under-studied. These languages are characterized as low-resource, since they usually lack the corresponding attention and/or data for the development of meaningful applications. Greek belongs to this language group. There is a dire need for the development of methods that will distill information from natural language content produced in Greek. This doctoral dissertation represents an attempt to meet the above need, through the design of a modern cognitive engine that enables the detection of latent correlations and patterns between entities, through the exploitation of the information wealth derived from Greek online sources and the combination of previously unrelated data (events, news, opinions etc.). This allows both the capture of information in a structured form, as well as its use for claim validation in natural language. More specifically, the dissertation utilizes automated crawling and pre-processing techniques on online news sources, in order to extract structured information that can be used for exploratory data analysis purposes and for the formulation of initial claims or hypotheses. In addition, it pertains to the development of advanced cognitive machine learning methods to achieve semantic inference and draw conclusions from the identification and connections between conceptual entities, ultimately aiming at the discovery of correlations between seemingly unrelated events, persons or actions. The final product of this work includes the design and implementation of a set of methodologies for information extraction and dynamic claim validation based on the accumulated information. All the above are accompanied by the development of corresponding machine learning models to support this work for the Greek use case. The mechanisms that will result from the development of the aforementioned methodologies allow the transformation of free-text to a structured representation (relational n-tuples), enabling better database management and enrichment with the help of external knowledge bases. Moreover, they render possible the validation or rejection of any textual claim, by aggregating heterogeneous information from multiple sources in real time, via a proposed evidence construction methodology.
Degree
thesis:*- Grantor dc:publisher
- Technical University of Crete (TUC)
- Year dc:date
- 2022
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Papadopoulos, Dimitrios
- Παπαδόπουλος, Δημήτριος
Subjects
dc:subject × 16- Επεξεργασία φυσικής γλώσσας
- Μηχανική μάθηση
- Βαθιά μάθηση
- Γλωσσικά μοντέλα
- Τεχνητή νοημοσύνη
- Εξαγωγή πληροφοριών
- Natural language processing
- Machine learning
- Deep learning
- Language models
- Artificial intelligence
- Information extraction
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Computer and Information Sciences
- Natural Sciences
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/52039