Bayesian Spectral Emotion Transition Discovery from Multi-Annotator Disagreement
Researchers have developed a new framework called Bayesian Spectral Emotion Transition Discovery (BSETD) to analyze how emotions shift during conversations. This method accounts for the uncertainty in multi-annotator judgments, unlike previous approaches that relied on majority voting. BSETD decomposes emotional transitions into components of inertia and contagion, revealing patterns such as the link between disgust and anger, and the under-representation of transitions from joy to anger. AI
IMPACT Provides a novel computational approach to understanding emotional dynamics in dialogue, potentially improving dialogue systems and mental-health screening tools.