This research reveals the specific mechanisms through which large language models process emotional content and make predictions about human emotions.

• Emotion representations are functionally localized to specific regions in the model architecture
• Findings validated across multiple model families and sizes with robust evaluation methods
• Research leverages cognitive appraisal theory to understand LLM emotion processing
• Identifies clear pathways for improving emotion inference capabilities

For the medical field, this work enables better mental health assessment tools, more empathetic therapeutic interfaces, and improved emotion-aware clinical decision support systems. Understanding these mechanisms is also crucial for developing ethical safeguards in emotion-sensitive applications.

Mechanistic Interpretability of Emotion Inference in Large Language Models