dc.contributor.author |
Awwad, Ramy Anwar, |
dc.date |
2014 |
dc.date.accessioned |
2015-02-03T10:35:03Z |
dc.date.available |
2015-02-03T10:35:03Z |
dc.date.issued |
2014 |
dc.date.submitted |
2014 |
dc.identifier.other |
b18262910 |
dc.identifier.uri |
http://hdl.handle.net/10938/10074 |
dc.description |
Thesis. M.E. American University of Beirut. Department of Electrical and Computer Engineering, 2014. ET:6021 |
dc.description |
Advisor : Dr. Hazem Hajj, Associate Professor, Electrical and Computer Engineering ; Members of Committee: Dr. Zaher Dawy, Associate Professor, Electrical and Computer Engineering ; Dr. Wassim El Hajj, Assistant Professor, Computer Science ; Dr. Fatima Al-Jamil, Assistant Professor, Psychology. |
dc.description |
Includes bibliographical references (leaves 45-51) |
dc.description.abstract |
Ability to feel emotions is known to be an intrinsic property of human beings. At every moment of our life, we unconsciously respond emotionally to everything that happens around us. Emotions are normally triggered by a person's environment and context, and then expressed through physiological changes and physical reactions. In this thesis we explore two modalities for emotion recognition: Text and Physiological. Among possible choices of context, text is the most common source of personal data. Physiological measurements, on the other hand, should present one of the common ways of showing emotional reactions. For emotion recognition from text, we propose a new method for optimal feature selection. We call this method MFX (the most frequent and discriminative features across). For physiological data, we conduct a study for improving accuracy of Ground truth data, and an evaluation for ranking of most relevant Physiological features. Experiment results with MFX show superiority for text classification with benchmark RCV1 and Reuters Data, and 85percent accuracy for emotion recognition from text. On the other hand, experiments with Physiological data, showed that raters' assessments provide more accurate ground truth data and higher recognition accuracies. Furthermore, the results of ranking physiological features indicate a correlation between certain features and certain aspects of emotions. For example, features extracted from Galvanic Skin Response (GSR) correlated more with Valence. |
dc.format.extent |
x, 51 leaves : illustrations (some color) ; 30 cm |
dc.language.iso |
eng |
dc.relation.ispartof |
Theses, Dissertations, and Projects |
dc.subject.classification |
ET:006021 AUBNO |
dc.subject.lcsh |
Machine learning. |
dc.subject.lcsh |
Artificial intelligence. |
dc.subject.lcsh |
Emotional intelligence. |
dc.subject.lcsh |
Data mining. |
dc.subject.lcsh |
Natural language processing. |
dc.subject.lcsh |
Support vector machines. |
dc.title |
Emotion recognition from text and physiological data - |
dc.type |
Thesis |
dc.contributor.department |
American University of Beirut. Faculty of Engineering and Architecture. Department of Electrical and Computer Engineering, degree granting institution. |