The aim of this thesis was to study the possibilities of utilizing affective information in human-computer interaction. Throughout the thesis, a dimensional theory of emotions has been used as a frame of reference. According to it, emotions can be defined as three dimensions, which are valence (negative-positive), arousal (calmhighly aroused), and dominance (controlledin control). The research is organized around four research goals. The first goal was to study, whether pupil size variation can be used in computer input to give the computer information about the user’s affective responses. The second goal was to study the effects of affective speech interventions to the user’s subjective experiences, physiology, and behavior. Third, we studied, whether experienced affective valence can be estimated reliably in real time by measuring the activations of the facial muscles associated with smiling and frowning. Fourth, the effects of the simulated proximity of a conversational character to the subjects’ affective experiences were studied. The results concerning pupil size variation showed generally that pupil size was bigger in response to positive and negative highly arousing stimuli, as compared to neutral stimuli. Thus, the results suggest that the magnitude of the pupillary response has a linear relationship with experienced affective arousal and a curvilinear relationship with experienced affective valence. The results concerning the experiment using affective interventions suggested that the users respond emotionally to human-computer interaction events and that by using affective interventions in the context of a negative event, positive affect-related changes can be induced in the user in terms of subjective experiences, facial activity, and task-related behavior. The results for the estimation of emotions based on the activations of facial muscles were approximately on the same level for both adaptive and person-independent estimation. The results supported the idea that affective valence can be estimated relatively reliably by measuring the activations of the muscles activated in smiling and frowning. The best estimation accuracies, discriminating between positive and negative responses at the accuracy of over 80%, were obtained by measuring frowning activity, and a difference score of the activations of the two muscles, when the subjects watched emotionally arousing videos. The results of the experiment, in which the proximity of a conversational agent was simulated by the size of the agent on the screen, suggested that the subjects’ experienced dominance or the feeling of control can be significantly influenced by varying the size of the agent on the screen. As a whole, the results suggest that pupil size variation and facial expressions can give the computer information about the user’s affective responses, and they could be potential input signals for human-computer interaction in the future. In addition, the results suggest that the wording of synthetic speech interventions and messages can be effectively used in computer output to influence the user’s affective physiological responses, behavior, and experienced valence. By varying the simulated proximity of a conversational agent, the users’ experienced affective dominance can also be influenced.
