OUSE: An Interactive Online User Satisfaction Evaluation Framework via Multimodal Emotional Intelligence

With the growth of online platforms, assessing user satisfaction is a challenge due to varying content and interaction experiences. In human-computer interaction, satisfaction includes emotional responses, cognitive evaluations, and behavioral engagement. To address limitations of multimodal affective computing methods focused on sentiment polarity, we propose Online User Satisfaction Evaluation (OUSE), a framework based on multimodal emotional intelligence. OUSE uses the Comprehensive Multimodal Opinion on Satisfaction (CMOS) dataset, annotated across affective, cognitive, and behavioral dimensions, and introduces Modality-Invariant and Specific Representations for Multimodal Sentiment Analysis with Satisfaction-awareness (MISA-S). MISA-S employs shared-private encoders with orthogonality constraints and Central Moment Discrepancy alignment to separate modality-specific nuances from shared semantics. A platform analyzes user feedback videos, generating evaluations of emotions, satisfaction, and behaviors. Experiments show MISA-S outperforms baselines, achieving R² for satisfaction prediction with 26.5%, 44.1%, and 46.3% improvements over MISA (R² = 0.452), MCTN (R² = 0.397), and MARN (R² = 0.391), providing a paradigm for multimodal satisfaction modeling.