Emergence of Hierarchical Emotion Organization in Large Language Models

The study analyzes the emergence of hierarchical emotion organization in large language models (LLMs), revealing that these models naturally construct emotion trees that align with human psychological frameworks. Larger models demonstrate more complex emotional hierarchies, but also exhibit systematic biases in emotion recognition, particularly for underrepresented socioeconomic groups. This research underscores the importance of incorporating cognitive theories in model evaluation to improve the ethical deployment of LLMs in conversational agents.