S3: Social-network Simulation System with Large Language Model-Empowered Agents

Chen Gao,Xiaochong Lan,Zhihong Lu,Jinzhu Mao,Jinghua Piao,Huandong Wang,Depeng Jin,Yong Li

2023-10-19

Abstract:Social network simulation plays a crucial role in addressing various challenges within social science. It offers extensive applications such as state prediction, phenomena explanation, and policy-making support, among others. In this work, we harness the formidable human-like capabilities exhibited by large language models (LLMs) in sensing, reasoning, and behaving, and utilize these qualities to construct the S$^3$ system (short for $\textbf{S}$ocial network $\textbf{S}$imulation $\textbf{S}$ystem). Adhering to the widely employed agent-based simulation paradigm, we employ prompt engineering and prompt tuning techniques to ensure that the agent's behavior closely emulates that of a genuine human within the social network. Specifically, we simulate three pivotal aspects: emotion, attitude, and interaction behaviors. By endowing the agent in the system with the ability to perceive the informational environment and emulate human actions, we observe the emergence of population-level phenomena, including the propagation of information, attitudes, and emotions. We conduct an evaluation encompassing two levels of simulation, employing real-world social network data. Encouragingly, the results demonstrate promising accuracy. This work represents an initial step in the realm of social network simulation empowered by LLM-based agents. We anticipate that our endeavors will serve as a source of inspiration for the development of simulation systems within, but not limited to, social science.

Social and Information Networks

What problem does this paper attempt to address?

The paper attempts to address the problem of using large language models (LLMs) to simulate user behavior in social networks, aiming to achieve accurate simulations of social phenomena such as information, attitude, and emotion dissemination. Specifically, the researchers have constructed a social network simulation system named S3, which, through fine-tuning and prompt engineering techniques, enables agents to mimic real human behavior in social networks, thereby observing group-level phenomena such as the spread of information, attitudes, and emotions.

### Main Contributions
1. **First application of large language models to social network simulation**: The researchers adopted an agent-based simulation paradigm and utilized the latest LLM technology to enhance the capabilities of the agents.
2. **Support for individual and group-level simulations**: The system can learn from actual social network data and simulate future states.
3. **High accuracy demonstrated in evaluation results**: The simulation system showed significant accuracy across multiple metrics, providing a new simulation paradigm for scientific research and practical applications in the social sciences.

### Research Background
- **Importance of social networks**: Social networks are a crucial part of modern society, and understanding their formation and evolution is vital for social scientists.
- **Application of simulation technology**: Simulation technology can predict future trends, support policy-making, and provide experimental environments.
- **Advantages of large language models**: LLMs possess powerful text understanding and generation capabilities, enabling them to simulate human behavior and decision-making processes.

### System Overview
- **Environment construction**: The environment is built using real social network data, and user demographic characteristics are inferred through prompt engineering and fine-tuning techniques.
- **Individual-level simulation**: By simulating users' emotions, attitudes, and interaction behaviors, the system evaluates their performance at the individual level.
- **Group-level simulation**: By accumulating individual behaviors, the system observes group-level phenomena of information, attitude, and emotion dissemination.

### Experiments and Evaluation
- **Experimental scenarios**: Controversial topics such as gender discrimination and nuclear energy were chosen for simulation.
- **Evaluation metrics**: Including accuracy, AUC, and F1 score, the results show that the system performs well across multiple tasks.

### Conclusion
This study demonstrates how large language models can be used to simulate user behavior in social network simulations, providing new tools and methods for social science research. Future work will further explore the potential of this field, especially in larger-scale and more application scenarios.