Abstract
Emotions widely affect the decision-making of humans and, hence, affectivecomputing takes emotional states into account with the goal of tailoringdecision support to individuals. However, the accurate recognition of emotionswithin narrative materials presents a challenging undertaking due to thecomplexity and ambiguity of language. Even though deep learning has evolved asthe state-of-the-art in various tasks from text mining, its benefits withregard to affective computing are not yet understood. We thus propose thefollowing innovations: (1) we adapt recurrent neural networks from the field ofdeep learning to affective computing. (2) We extend these networks forpredicting the score of different affective dimensions. (3) We implementtransfer learning for pre-training word embeddings. Analyzing the results, wefind that deep learning consistently outperforms traditional machine learningwith improvements of up to 21% in F1-score when labeling emotions and 6% inforecast errors when rating the intensity of emotions. Altogether, the findingshave considerable implications for the use of affective computing in providingdecision support.