Social Opinions Prediction Utilizes Fusing Dynamics Equation with LLM-based Agents

TL;DR

The result demonstrates that the FDE-LLM significantly outperforms traditional Agent-Based Modeling (ABM) algorithms and LLM-based algorithms and accurately simulates the decay and recovery of opinions over time, underscoring LLMs potential to revolutionize the understanding of social media dynamics.

Abstract

In the context where social media emerges as a pivotal platform for social movements and shaping public opinion, accurately simulating and predicting the dynamics of user opinions is of significant importance. Such insights are vital for understanding social phenomena, informing policy decisions, and guiding public opinion. Unfortunately, traditional algorithms based on idealized models and disregarding social data often fail to capture the complexity and nuance of real-world social interactions. This study proposes the Fusing Dynamics Equation-Large Language Model (FDE-LLM) algorithm. This innovative approach aligns the actions and evolution of opinions in Large Language Models (LLMs) with the real-world data on social networks. The FDE-LLM devides users into two roles: opinion leaders and followers. Opinion leaders use LLM for role-playing and employ Cellular Automata(CA) to constrain opinion changes. In contrast, opinion followers are integrated into a dynamic system that combines the CA model with the Susceptible-Infectious-Recovered (SIR) model. This innovative design significantly improves the accuracy of the simulation. Our experiments utilized four real-world datasets from Weibo. The result demonstrates that the FDE-LLM significantly outperforms traditional Agent-Based Modeling (ABM) algorithms and LLM-based algorithms. Additionally, our algorithm accurately simulates the decay and recovery of opinions over time, underscoring LLMs potential to revolutionize the understanding of social media dynamics.

Figures (8)

• Figure 1: Event Relationship Network Diagram. The core of the network represents opinion leaders, while the branches represent opinion followers. The deeper the color, the stronger the influence.
• Figure 2: The red lines represent actual data (a) Pangmao Incident (b) Jiangping Incident (c) Qingdao Incident (d) Dianduji Incident. The labeled parts in the figures are the key areas that we need to simulate using the models.
• Figure 3: Workflow of the FDE-LLM Simulation: Cooperation of Opinion Leaders and Opinion Followers, Role-Playing Mechanism, and Integration of CA and SIR Models.
• Figure 4: Actions and Attitudes. The left side displays the types of actions that LLM-Action can select. In contrast, the right side illustrates the specific behaviors related to each action type and the associated scoring of attitudes by LLM-Attitudes.
• Figure 5: Agent Profile. This is an example of an active and influential social media agent with narrative amplification and exaggerated commentary to engage the audience. The profile is summarized from the actual dataset, based on the target user's reviews and posts