From Valence to Emotions

The thesis explores the potential of user-generated content for predicting human-related events by analyzing the emotions and valences expressed within it. It provides a theoretical framework for sentiment detection and classification methods, categorizing empirical literature into three prediction subjects. A comprehensive analysis reveals significant differences in prediction accuracies based on data sources and methods, highlighting that fine-grained sentiments enhance prediction accuracy. The study also addresses the scarcity of empirical data on fine-grained sentiment approaches for evaluation.