ConceptSelective Deviation Prediction
Selective Deviation Prediction
Selective deviation prediction is a task framework for conversational emotion modeling. Rather than continuously predicting emotion values, it focuses on determining when emotion will deviate from a persistence baseline and issues an alert only at that moment. The framework typically combines conformal risk control to provide statistical guarantees for committed predictions, and suits scenarios requiring highly reliable alerts.
Timeline (last 90 days)
Nov 5
The study reframes the conversational emotion prediction task as selective deviation prediction: predicting only when emotion will deviate from the persistence baseline, rather than predicting specific emotion values.
Unverified50%