OneEmo: A Unified Multimodal Reasoning Model for Emotion Perception, Understanding, and Interaction

2026-08-06

AI Summary

This work addresses the limitation of existing affective intelligence models, which are typically confined to single tasks and thus fail to exploit cross-task synergies. To overcome this, we propose OneEmo, a unified general-purpose emotional intelligence model that integrates emotion perception, understanding, and interaction through multi-task joint learning. Our key contributions include the construction of EmoWorld-130K, a large-scale multi-task dataset; the design of Emo-Chord, a reinforcement learning strategy that explicitly shares reasoning trajectories across tasks; and a unified optimization framework combining human feedback distillation, supervised fine-tuning, and a shared reward mechanism. Experimental results demonstrate that OneEmo outperforms same-scale baselines on most benchmarks and achieves performance comparable to commercial systems with significantly fewer parameters.

Abstract

Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities in emotional intelligence. However, prevailing research predominantly focuses on task-specific specialization, often neglecting inter-task synergy and leaving latent reasoning potential underexplored. To bridge this gap, we introduce OneEmo, a unified affective generalist capable of mastering emotion perception, comprehension, and interaction. For this purpose, we first construct EmoWorld-130K, a comprehensive dataset that distills specialized affective knowledge into explicit reasoning trajectories via a human-in-the-loop workflow. Supervised fine-tuning on this corpus reveals significant mutual benefits derived from multi-task learning. Second, to fully unlock the latent reasoning potential, we propose Emo-Chord, a novel reinforcement learning strategy that stabilizes optimization through unified multi-task reward allocation. Extensive experiments demonstrate that OneEmo achieves state-of-the-art performance against similarly sized baselines across most benchmarks. Notably, despite having significantly fewer parameters than commercial models, OneEmo delivers highly competitive results. This paper paves the way for more reliable and interpretable affective computing. The code is available at https://github.com/waHAHJIAHAO/OneEmo.

Problem

• Multimodal Large Language Models
• Emotion Perception
• Emotion Understanding
• Emotion Interaction
• Affective Computing

Innovation

• unified multimodal reasoning
• affective generalist
• EmoWorld-130K
• Emo-Chord
• multi-task reinforcement learning