EmoScene

Emotion Entanglement and Bayesian Inference for Multi-Dimensional Emotion Understanding

Announce Type: replace Abstract: Understanding emotions in natural language is inherently a multi-dimensional reasoning problem, where multiple affective signals interact through context, interpersonal relations, and situational cues. However, most existing emotion understanding benchmarks rely on short texts and predefined emotion labels, reducing this process to independent label prediction and ignoring the structured dependencies among emotions. To address this limitation, we introduce...