Human-Computer Interaction migrates from the classic perspective to a more natural environment, where humans are able to use natural language to exchange knowledge with a computer. In order to fully “understand” the human’s intentions, the computer should be able to detect emotions and reply accordingly. This thesis focuses on several issues regarding the human affects, from various detection techniques to their integration into a Distributed Interactive System. Emotions are a fuzzy concept and their perception across human individuals may vary as well. Therefore, this makes the detection problem very difficult for a computer. From the affect detection perspective, we proposed three different approaches: an emotion detection method based on Self Organizing Maps, a valence classifier based on multi-modal features and Support Vector Machines, and a technique to resolve conflicts into a well known affective dictionary (SentiWordNet). Moreover, from the system integration perspective, two issues are approached: a Wizard of Oz experiment in a children storytelling environment and an architecture for a Distributed Interactive System.
