| 16 |
Capturing subjective feelings objectivelyModeling emotions with response styles and uncertainty 
|
|---|
Understanding others' emotions is crucial to human communication. However, trying to understand emotions through subjective questions is often hindered by individual response styles and subjective uncertainties. By mathematically modeling these response styles and uncertainties, we approach subjective emotions objectively. By statistically analyzing responses to various subjective questions, we detect individual response styles and the uncertainties in answers that vary on different occasions, even when the same question is asked repeatedly. Whether someone is overly expressive, reserved, or unclear in their emotional recognition, we aim to unveil their true emotions behind their responses. Human communication is complex, and emotional expression and cognition vary among individuals. By creating AI that comprehends human emotions, including response styles and uncertainties, we aim to reduce misunderstandings caused by unclear communication and foster a more empathetic society.
[1] S. Kumano, K. Nomura, “Multitask item response models for response bias removal from affective ratings,” in Proc. The 8th International Conference on Affective Computing and Intelligent Interaction (ACII ’19), 2019.
[2] H. Narimatsu, M. Ozawa, S. Kumano, “Collision probability matching loss for disentangling epistemic uncertainty from aleatoric uncertainty,” in Proc. The 26th International Conference on Artificial Intelligence and Statistics (AISTATS ’23), PMLR 206:11355-11370, 2023.
Shiro Kumano, Sensory Resonance Research Group, Human Information Science Laboratory