Abstract

Different theories of emotions have been introduced since the 19th century. Even though a large number of apparent differences between these theories exist, there is a broad consensus today that emotions consist of multiple components such as cognition, physiology, motivation, and subjectively perceived feeling. Appraisal theories of emotions, such as the Component Process Model (CPM) by Klaus Scherer, emphasize that the cognitive evaluation of a stimulus or event is the driving component of the emotion process. It is believed to cause changes in all other components and hence to differentiate emotion states. To test the CPM and gain more insights into the multi-componential emotion process, the present thesis examines two emotion sub-processes – the link between the cognitive and the feeling component (study 1) and the link between the cognitive and the physiological component (study 2) – by using different predictive modeling approaches. In study 1, four theoretically informed models were implemented. The models use a weighted distance metric based on an emotion prototype approach to predict the perceived emotion of participants from self-reported cognitive appraisals. Moreover, they incorporate different weighting functions with weighting parameters that were either derived from theory or estimated from empirical data. The results substantiate the examined link based on the predictive performance of the models. In line with the CPM, the preferred model weighted the appraisal evaluations differently in the distance metric. However, the data-derived weighting parameters of this model deviate from theoretically proposed ones. Study 2 analyzed the link between cognition and physiology by predicting self-reported appraisal dimensions from a large set of physiological features (calculated from different physiological responses to emotional videos) using different linear and non-linear machine learning algorithms. Based on the predictive performance of the models, the study is able to confirm that most cognitive evaluations were interlinked with different physiological responses. The comparison of the different algorithms and the application of methods for interpretable machine learning showed that the relation between these two components is best represented by a non-linear model and that the studied link seems to vary among physiological signals and cognitive dimensions. Both studies substantiate the assumption that the cognitive appraisal process is interlinked with physiology and subjective feelings, accentuating the relevance of cognition in emotion as assumed in appraisal theory. They also demonstrate how computational emotion modeling can be applied in basic research on emotions.