NPC-Valence-Arousal-Prediction-ONNX

NPC Valence Arousal Prediction

ONNX version of : Mavdol/NPC-Valence-Arousal-Prediction

What is it?

This model predicts emotional states in video game NPC dialogues using Russell's Circumplex Model of Affect. Trained on dialogue datasets from popular RPGs including Skyrim, Baldur's Gate, Cyberpunk 2077, and others, it analyzes text input and outputs two continuous emotional dimensions:

• Valence: How positive or negative the emotion feels (ranging from unpleasant to pleasant)
• Arousal: How intense or calm the emotion is (ranging from low energy to high energy)

Objective: Enable more nuanced emotional understanding in gaming contexts by moving beyond discrete emotion categories to a continuous 2D emotional space that better captures the complexity of NPC emotional expressions.

Russell's Circumplex Model - Core Principles

The Circumplex Model represents emotions as points in a circular 2D space defined by two independent axes:

• Valence (X-axis): Pleasant ↔ Unpleasant
  • Positive values = positive emotions (joy, contentment)
  • Negative values = negative emotions (sadness, anger)
• Arousal (Y-axis): High Activation ↔ Low Activation
  • High values = energetic emotions (excitement, rage)
  • Low values = calm emotions (relaxation, depression)

Why This Approach?

Unlike traditional emotion classification that uses fixed categories, the circumplex model:

• Captures emotional intensity and nuance
• Allows for smooth transitions between emotional states
• Better represents the continuous nature of human emotions
• Provides context-appropriate emotional understanding for gaming scenarios

Emotions are plotted as coordinates in this circular space.

🎮 Try the Interactive Valence-Arousal Visualizer - Click anywhere in the circle to explore different emotional coordinates!

Citations

@dataset{NPC-Valence-Arousal-Prediction-ONNX,
  title={Valence and Arousal Annotations for Interactive Characters},
  author={Mavdol},
  year={2025},
  url={https://huggingface.co/Mavdol/NPC-Valence-Arousal-Prediction-ONNX},
  note={Based on Russell's Circumplex Model of Affect for NPC emotion recognition}
}