Abstract
dc:descriptionRecent years have witnessed a number of important efforts aimed at technologies for automatic detection and analysis of human emotion using physiological, facial and voice signals, as well as personal interpretations from social media texts. There are a number of research areas that could benefit from such systems: interactive teaching systems which allow teachers to identify student stress, accident prevention (e.g. driver's fatigue detection), military tasks which are characterized by long stress and pressure periods and health care applications for timely diagnosis of neurodegenerative diseases (e.g., Parkinson's disease), where the manifestation of symptoms occurs many years after the onset of neurodegeneration.However, despite the research efforts so far, the long-term objective of creating a strong framework concerning the examined research field that should be based on its analysis and interpretation has not been achieved so far. There is no no doubt that ``affect production" is affected from the context that is taking place at the moment, such as the resources that are nearby, What (the human is doing), the situation that is occurring, the identities of the people around the user as well as the degree of expressiveness. Any other complementary contextual information examined assists us to respond to the question: what is more likely to happen, thus diverting the classifier from more likely/relevant categories. Without the framework, even people can misinterpret its observable expressions. Thus, by addressing the above described challenges under the context of analyzing emotion in a specific context (context-aware affect analysis), i.e. studying the contextual parameters of the framework through its interpretation in specific areas of application, its representation, its modeling, we may better approach context analysis in real time.In the field of personal interpretations from the text (Sentiment Analysis) and more generally in the field of Natural Language Processing (NLP), the contribution of the framework lies in a more effective analysis, interpretation and processing of opinions and sentiments in text which are examined at document, sentence and aspect level respectively. In this case, the semantic, cognitive and emotional information of the subjective responses of individuals are taken into account. To be more precise, in this area, the our contribution is based on the idea of training strong feature representations from unlabelled data using Neural Networks and particularly using the Generative Adversarial Neural Networks (GANs), which have proved very popular in the Computer Vision domain. The originality of our proposed method lies in the model's implementation, in the hyper-parameters' choice, in the use of unsupervised learning and finally in its experimental validation on text bodies which come from different sources depending on their genre and length.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2018
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Vlachostergiou, Aggeliki-Spyridoula
- Βλαχοστέργιου, Αγγελική-Σπυριδούλα
Subjects
dc:subject × 10- Ανάλυση συναισθήματος
- Ανάλυση γνώμης
- Συναισθηματική υπολογιστική
- Emotion analysis
- Sentiment analysis
- Opinion mining
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Engineering and Technology
- Electrical Engineering, Electronic Engineering, Information Engineering
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/44752