EmoGavel improves emotion preference learning with error correction
EmoGavel: Grounding-Aware Reinforcement Learning with Verifiable Rewards for Multimodal Emotion Preference Learning
MultimediaMachine Learning
Summary
Understanding human emotions is complicated because different people describe feelings in many ways. This paper introduces EmoGavel, a new method that not only guesses which emotion descriptions people prefer but also finds and fixes mistakes in those guesses. The method trains itself by learning from examples with known errors and improves how it judges emotions by using special rewards that check for correctness. Tests show that EmoGavel is better at predicting preferences and spotting errors than previous methods. The approach helps make emotion understanding from multiple types of data more accurate and trustworthy.
What this means in practice
- •For conversational ai developers: Enhance chatbot responses by improving how systems learn and verify emotional preferences from multiple data sources.$Commercial implications: Enables creation of emotionally aware chatbots with more accurate and verifiable emotion recognition, improving user experience.
- •For multimodal data engineers: Improve models that integrate text, audio, and visual signals to better identify and correct emotion preference errors during training.
Authors
Zilong Huang, Kong Aik Lee, Zezhong Jin, Man-Wai Mak, Tatsuya Kawahara
Abstract
Multimodal emotion preference learning uses pairwise comparisons to identify which open-ended emotion descriptions align more closely with human judgments. However, existing methods primarily focus on the final preference decision and provide little explicit supervision for localizing and correcting erroneous content. To address this limitation, we propose EmoGavel, a Grounding-Aware reinforcement learning framework with Verifiable rewards for multimodal Emotion preference Learning. We first introduce grounded supervised fine-tuning (Grounded SFT), which exploits controlled-error construction information to jointly learn preference prediction, error localization, repair, and missing-evidence recovery. We then optimize these verifiable outcomes with grounding-aware group relative policy optimization (GA-GRPO), combining preference-gated grounding rewards with dynamic zero-variance rescue to prioritize preference correctness and recover learning signals from all-wrong groups. Experiments on EmoPrefer and controlled-error evaluation sets show that EmoGavel improves both natural preference prediction and controlled-error discrimination while maintaining strong error-grounding performance. Our code and model checkpoints will be released once the paper is accepted.