University of Patras
Μηχανική μάθηση και επεξεργασία φυσικής γλώσσας: σημασιολογικές αναπαραστάσεις, ανάλυση συναισθήματος και μοντελοποίηση χρονοσειρών
Abstract
dc:descriptionThis work addresses, firstly, the application and proposal of diverse frameworks for representing emotional states and attitudes in time series modeling through ensembles, and secondly, the conceptual development and theoretical exploration of semantic content embeddings. Through the results of an extensive experimental process focused on pandemic, financial, and energy-related time series data, we empirically demonstrate that such time series phenomena are subtly shaped by linguistic behaviors as articulated within various human communities. Based on these findings, we introduce a series of specific algorithmic structures. The proposed architectures integrate multivariate feature spaces informed by diverse emotion modeling approaches. These models are initially derived from specialized thematic content and then extended through general linguistic frameworks. For each dataset, we suggest a corresponding series of ensemble schemes. Within this context, we show that numerous linguistic contents embed, through emotional modeling, information pertinent to the specific time series phenomena under study. Additionally, we identify underlying patterns in the sequencing of emotional states and attitudes across various linguistic data streams. These patterns are then modeled to extract the relevant underlying latent information, operating under the assumption of a distributional structure established by a general methodology that we propose. Our findings suggest that this distributional structure for embedding emotional attitudes not only enhances models tackling natural language processing challenges but also improves predictive frameworks for the specific time series data under consideration. Consequently, the representations offer information that directly contributes to the mathematical description of the studied phenomena. Lastly, we demonstrate that the corresponding representational frameworks—whether purely semantic or emotion-based—capture semantic content through object relationships grounded in syntactic forms. This approach thus establishes a metric-based, end-to-end framework.
Degree
thesis:*- Grantor dc:publisher
- University of Patras
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Liapis, Charalampos
- Λιάπης, Χαράλαμπος
Subjects
dc:subject × 28- Μηχανική μάθηση
- Επεξεργασία φυσικής γλώσσας
- Μοντελοποίηση χρονοσειρών
- Ανάλυση συναισθήματος
- Πολυμεταβλητή ανάλυση
- Ομαδοποιητές
- Ενσωματώσεις
- Κατανεμημένες ενσωματώσεις συναισθημάτων
- Σημασιολογικό περιεχόμενο
- Machine learning
- Natural language processing
- Time series modeling
- Sentiment analysis
- Multivariate analysis
- Ensemble learning
- Embeddings
- Distributional emotion embeddings
- Semantic content
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Τεχνητή νοημοσύνη
- Μαθηματικά
- Υπολογιστικά μαθηματικά
- Computer and Information Sciences
- Natural Sciences
- Artificial Intelligence
- Mathematics
- Computational Mathematics
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/58668