Selective Departure Forecasting for Conversational Emotion Prediction

The study reframes emotion forecasting from predicting continuous valence-arousal-dominance states to detecting departures from a speaker's persistence baseline. Using lead-time-weighted metrics and conformal guarantees, the authors propose a deployment-ready framework for predicting when emotional trajectories will shift.