Eye-Tracking Based Identification of Visual Pathways in Emotional Watercolor Narratives

Affective watercolor narratives express affective information using the techniques of color diffusion, hierarchical composition, positioning of characters, and ordered scene presentation. However, there still lacks sufficient quantifiable representations of the attention transitions that play the roles during their perception. Eye movements of 72 participants were tracked by a 120Hz eye tracker while they were watching 30 positive, negative, and neutral watercolor narratives. Fixation, saccade, revisit, and transition between AOIs were extracted. XGBoost was used for emotion classification task while HMM algorithm identified latent scanpath states. In this particular dataset, 2086 valid trials were selected for analysis, which constitutes a retention rate of 96.6%. In terms of attentional transitions, negative narratives induced significantly shorter time to first fixation within relevant AOIs (498 ± 156 ms), longer fixation duration of 1715 ± 455 ms and revisit rate of 0.44 ± 0.16, which is indicative of attentional capture. XGBoost reached a Macro F1 of 0.81 and a macro AUC of 0.91, which outperformed the Logistic Regression baseline model of 0.68 and 0.82. HMM analysis results demonstrated a strong local loop between Face and Key Object regions in negative cases while more dispersed transitions toward background regions were observed in positive narratives. Integration of both gaze based classification and sequential scanpath modeling gives an interpretable computational representation of emotional visual attention.