CHEER-Ekman: Fine-grained Embodied Emotion Classification

2025-06-01

AI Summary

This work addresses two key challenges in embodied emotion recognition—namely, the difficulty of identifying emotions grounded in bodily sensations and physical experiences, and the limitation of existing datasets to coarse-grained categories (e.g., Ekman’s six basic emotions). To this end, we introduce CHEER-Ekman, the first fine-grained embodied emotion classification dataset. Methodologically, we extend embodied emotion modeling to a six-class framework for the first time and propose a large language model (LLM)-based Best-Worst Scaling (BWS) annotation paradigm. We further demonstrate that simplified prompting and chain-of-thought (CoT) reasoning significantly enhance small-model performance. Experiments show that CHEER-Ekman effectively supports embodied emotion modeling; our approach outperforms supervised learning baselines on the new benchmark; and optimized small models achieve accuracy comparable to LLMs—offering an efficient, resource-conscious alternative for deployment-constrained scenarios.

Technology Category

• Intelligent Robots: Embodied AI
• Humans and AI: Emotional Intelligence
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Emotions manifest through physical experiences and bodily reactions, yet identifying such embodied emotions in text remains understudied. We present an embodied emotion classification dataset, CHEER-Ekman, extending the existing binary embodied emotion dataset with Ekman's six basic emotion categories. Using automatic best-worst scaling with large language models, we achieve performance superior to supervised approaches on our new dataset. Our investigation reveals that simplified prompting instructions and chain-of-thought reasoning significantly improve emotion recognition accuracy, enabling smaller models to achieve competitive performance with larger ones.

Problem

• Identifying embodied emotions in text is understudied
• Extending binary emotion dataset with six basic categories
• Improving emotion recognition accuracy with simplified prompting

Innovation

• Extends dataset with Ekman's six emotion categories
• Uses automatic best-worst scaling with LLMs
• Simplified prompting and chain-of-thought improve accuracy