Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Large Emotional World Model

Changhao Song, Yazhou Zhang, Hui Gao, Chang Yang, Peng Zhang

TL;DR

This work identifies a gap in traditional world models: they underrepresent emotional factors that drive real-world behavior. It proposes the Emotion-Why-How (EWH) dataset and the Large Emotional World Model (LEWM), which jointly model emotional states, perceptual observations, and actions to predict future states and emotional transitions. Empirical results show that emotional signals can modulate model behavior, with LEWM achieving improved accuracy in predicting emotion-driven social actions while maintaining competitive performance on basic tasks. The approach advances human-aligned world modeling by explicitly incorporating affective reasoning into latent dynamics, offering practical benefits for social simulation and interactive AI systems.

Abstract

World Models serve as tools for understanding the current state of the world and predicting its future dynamics, with broad application potential across numerous fields. As a key component of world knowledge, emotion significantly influences human decision-making. While existing Large Language Models (LLMs) have shown preliminary capability in capturing world knowledge, they primarily focus on modeling physical-world regularities and lack systematic exploration of emotional factors. In this paper, we first demonstrate the importance of emotion in understanding the world by showing that removing emotionally relevant information degrades reasoning performance. Inspired by theory of mind, we further propose a Large Emotional World Model (LEWM). Specifically, we construct the Emotion-Why-How (EWH) dataset, which integrates emotion into causal relationships and enables reasoning about why actions occur and how emotions drive future world states. Based on this dataset, LEWM explicitly models emotional states alongside visual observations and actions, allowing the world model to predict both future states and emotional transitions. Experimental results show that LEWM more accurately predicts emotion-driven social behaviors while maintaining comparable performance to general world models on basic tasks.

Large Emotional World Model

TL;DR

This work identifies a gap in traditional world models: they underrepresent emotional factors that drive real-world behavior. It proposes the Emotion-Why-How (EWH) dataset and the Large Emotional World Model (LEWM), which jointly model emotional states, perceptual observations, and actions to predict future states and emotional transitions. Empirical results show that emotional signals can modulate model behavior, with LEWM achieving improved accuracy in predicting emotion-driven social actions while maintaining competitive performance on basic tasks. The approach advances human-aligned world modeling by explicitly incorporating affective reasoning into latent dynamics, offering practical benefits for social simulation and interactive AI systems.

Abstract

World Models serve as tools for understanding the current state of the world and predicting its future dynamics, with broad application potential across numerous fields. As a key component of world knowledge, emotion significantly influences human decision-making. While existing Large Language Models (LLMs) have shown preliminary capability in capturing world knowledge, they primarily focus on modeling physical-world regularities and lack systematic exploration of emotional factors. In this paper, we first demonstrate the importance of emotion in understanding the world by showing that removing emotionally relevant information degrades reasoning performance. Inspired by theory of mind, we further propose a Large Emotional World Model (LEWM). Specifically, we construct the Emotion-Why-How (EWH) dataset, which integrates emotion into causal relationships and enables reasoning about why actions occur and how emotions drive future world states. Based on this dataset, LEWM explicitly models emotional states alongside visual observations and actions, allowing the world model to predict both future states and emotional transitions. Experimental results show that LEWM more accurately predicts emotion-driven social behaviors while maintaining comparable performance to general world models on basic tasks.
Paper Structure (10 sections, 15 equations, 5 figures)

This paper contains 10 sections, 15 equations, 5 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. Preliminary
  4. Methodology
  5. Emotion-Why-How(EWH) Dataset
  6. Large Emotional World Model(LEWM)
  7. Experiment
  8. Validation Results
  9. Conclusion
  10. Example Appendix

Figures (5)

  • Figure 1: World model based solely on physics regularities fails to predict emotion-driven behavior.
  • Figure 2: The world model integrated with an emotion-filtering module is trained using a two-stage approach.
  • Figure 3: Overview of EWH Dataset.
  • Figure 4: Illustrations of LEWM model architecture
  • Figure 5: Model performance across different tasks.