Affective computing (AC) is an emerging multidisciplinary field, which aims to bridge the gap between the highly emotional human and the emotionally challenged computer, in order to provide a higher level of human machine interaction. This Thesis contributes to AC research by proposing new emotion representations, and by developing novel computational mechanisms and methodological frameworks to be utilised by AC systems. This research is conducted under an educational scope, however it is not limited to an educational context since the methodologies presented can be applied in a number of different areas, thus offering a large potential of future research directions. The contributions of this research fall under the scope of AC, machine learning, and the psychological theories aiming to understand human emotion. A contextualized and personalised version of the Affective Trajectories hypothesis is presented, extending on the original theory, and a framework for its utilization by AC systems is provided. A fuzzy mechanism for knowledge extraction and adaptation is developed, which achieved an improved classification performance compared to the computational methods it relied upon. A novel computational model of emotion, the AV-AT model, is proposed, providing AC researchers with a tool to describe efficiently user's affective state. A custom hierarchical fuzzy method is developed, consisting of a genetically optimized adaptive fuzzy system, and a Fuzzy Cognitive Map. This method incorporates low-level information concerning the basic elements of a student's affective trajectory through time, and high-level information of the affective transitions a student experiences, to model students' affective trajectories during learning tasks. The affective transitions of students are explored in the context of problem based learning pedagogical frameworks. A novel scenario based survey design is introduced to elicit affect information. Finally, an offline adaptation process is presented, to enable the development of pre-trained personalised systems.
|Date of Award||Apr 2017|
|Supervisor||Faiyaz Doctor (Supervisor), Rahat Iqbal (Supervisor) & Anne James (Supervisor)|