Synthetic Social Media Influence Experimentation via an Agentic Reinforcement Learning Large Language Model Bot

TL;DR

The paper tackles the challenge of ethically studying public opinion dynamics on social platforms by building a simulated environment that couples agentic reinforcement learning with large language models. It trains a target agent to maximize followers using Q-learning, while LLMs convert RL decisions into topic-relevant posts. Key contributions include a topic-aware ABM framework, empirical evidence that constraining actions and enabling self-observation yield more stable opinion-leader formation, and analysis across multiple observability and model-settings scenarios. The work has practical significance for understanding influence mechanisms, informing policy and marketing strategies, and providing a controllable testbed to explore misinformation risks and detection in online ecosystems.

Abstract

Understanding the dynamics of public opinion evolution on online social platforms is crucial for understanding influence mechanisms and the provenance of information. Traditional influence analysis is typically divided into qualitative assessments of personal attributes (e.g., psychology of influence) and quantitative evaluations of influence power mechanisms (e.g., social network analysis). One challenge faced by researchers is the ethics of real-world experimentation and the lack of social influence data. In this study, we provide a novel simulated environment that combines agentic intelligence with Large Language Models (LLMs) to test topic-specific influence mechanisms ethically. Our framework contains agents that generate posts, form opinions on specific topics, and socially follow/unfollow each other based on the outcome of discussions. This simulation allows researchers to observe the evolution of how opinions form and how influence leaders emerge. Using our own framework, we design an opinion leader that utilizes Reinforcement Learning (RL) to adapt its linguistic interaction with the community to maximize its influence and followers over time. Our current findings reveal that constraining the action space and incorporating self-observation are key factors for achieving stable and consistent opinion leader generation for topic-specific influence. This demonstrates the simulation framework's capacity to create agents that can adapt to complex and unpredictable social dynamics. The work is important in an age of increasing online influence on social attitudes and emerging technologies.

Figures (7)

• Figure 1: (a) Simulated Environment Social Network: Participants within this environment, referred to as 'agents', are interconnected through the 'follow' relationship between each other. Agents interact through 'Posts' which is a small natural language message, formatted like a Tweet. Social influence evolution process shows the development of the whole network from $t$ to $t+1$. The process can be segmented into three part: Post Generation, Opinion Elicitation and Link (follow) Update. This process is how we track and measure influence. (b) Agent Design Principle: The agent is comprised of 2 building blocks - (i) the RL model that formulates how best to strategically gain influence in the long term, and (ii) the LLM that maps the strategy decisions from RL to natural language tweets for the topic specified.
• Figure 2: RL Loop. Reward: change of number of followers at each time step. State: 1)the current opinion states of the initial accounts followed by the target agent, 2)the number of followers the target agent currently has. Action: pre-defined five opinion polarity categories.
• Figure 3: Distribution of polarity value.
• Figure 4: Opinion Evolution Visualization on Narrow setting and Creative setting
• Figure 5: Standard Deviation of Opinion on Narrow setting and Creative setting