Applying appraisal to detect emotions in a real-world, multi-tasking empathic space

• Main Author: Chuacokiong, Marjorie Wee
• Format: text
• Language: English
• Published: Animo Repository 2012
• Subjects: Emotion recognition; Emotions
• Online Access: https://animorepository.dlsu.edu.ph/etd_masteral/4273
• Institution: De La Salle University

This paper investigates the use of appraisals in a real-world, multi-tasking environment such as an empathic space. While several appraisal models exist which prove successful in determining a persons affect, they have not been applied to a real-world living environment where emotion is highly influenced by an individuals goals and activities. In this research, a computational model of appraisal is proposed which makes use of knowledge-driven rules and data-centric probabilities to map activities into two-dimensional arousal-valence values. Specifically, affect is derived using seven appraisal variables from Scherer which are suddenness, intrinsic pleasant- ness, goal relevance, unpredictability, outcome probability, discrepancy from expectation, and goal conduciveness. Evaluations were conducted based on the corpus built with 4 occupants engaged in their normal daily activities. The system gained an average correlation coefficient (CC) of 0.7302 and a root mean squared error (RMSE) of 0.0575 for predicting arousal, while it yielded a perfect CC and RMSE for valence. Further experiments revealed the high correlation of the appraisal variables suddenness and intrinsic pleasantness to the arousal dimension.