Agent-Based Modelling Meets Generative AI in Social Network Simulations

TL;DR

A novel framework utilizing LLM-empowered agents to simulate social network users based on their interests and personality traits is proposed, demonstrating that LLM-agents accurately replicate real users' behaviors, including linguistic patterns and political inclinations.

Abstract

Agent-Based Modelling (ABM) has emerged as an essential tool for simulating social networks, encompassing diverse phenomena such as information dissemination, influence dynamics, and community formation. However, manually configuring varied agent interactions and information flow dynamics poses challenges, often resulting in oversimplified models that lack real-world generalizability. Integrating modern Large Language Models (LLMs) with ABM presents a promising avenue to address these challenges and enhance simulation fidelity, leveraging LLMs' human-like capabilities in sensing, reasoning, and behavior. In this paper, we propose a novel framework utilizing LLM-empowered agents to simulate social network users based on their interests and personality traits. The framework allows for customizable agent interactions resembling various social network platforms, including mechanisms for content resharing and personalized recommendations. We validate our framework using a comprehensive Twitter dataset from the 2020 US election, demonstrating that LLM-agents accurately replicate real users' behaviors, including linguistic patterns and political inclinations. These agents form homogeneous ideological clusters and retain the main themes of their community. Notably, preference-based recommendations significantly influence agent behavior, promoting increased engagement, network homophily and the formation of echo chambers. Overall, our findings underscore the potential of LLM-agents in advancing social media simulations and unraveling intricate online dynamics.

Figures (6)

• Figure 1: Our framework comprises two primary phases: (i) Characterization, where each agent embodies the personality traits and interests extracted (via LLM) from the original posts of the real user it is tasked to emulate; and (ii) Simulation, where the decision-making process of each agent, represented as a Choice-Reason-Content triple (Reasoning Module), is stored within the Interaction Module. Consequently, each agent autonomously makes decisions, considering the context and having access to recommended contents posted by other agents.
• Figure 2: Workflow of the RAG-empowered interaction: At the beginning of the Simulation phase, the first agent's decision is stored in the vector database. The RAG mechanism of the Interaction Module retrieves contextually relevant data from the database to enrich the prompt for the next agent to assist in making informed decisions within the Reasoning Module, taking into account the current environmental state and previously published content by other agents. The retrieval-informed decision is, in turn, stored in the vector database, repeating the cycle until the simulation ends.
• Figure 3: The distributions of keywords usage in real Twitter discussions and the simulation. The x-axis represents the number of occurrences of a keyword, while the y-axis represents the probability of encountering a keyword with a specific frequency.
• Figure 4: Comparison of self-similarity distributions between real users and LLM-agents.
• Figure 5: Comparison of intra-cluster similarity distributions between real users and LLM-agents.