Motivation: Recent global conflicts, notably the Ukraine war and Israel-Hamas tensions, underscore the pivotal role of social media in shaping narratives, disseminating information, and uplifting morale. Understanding the impact of biased reporting during warfare is crucial, as it molds public perception and influences the course of events. Additionally, social media serves as a valuable resource for defense personnel in refining targeting strategies.

Problem: Military organizations face legal constraints in collecting and analyzing social media data. Evaluating the consequences of actions during conflicts through social media analysis is limited to real-time or retrospective assessments. The absence of predictive capabilities impedes training, contingency planning, and the formulation of guidelines for extracting information during conflicts.

Method: Building upon prior research using simulation techniques to replicate social behavior, this study integrates external events into simulated communities. We generate realistic social media populations and interactions by leveraging state-of-the-art Large Language Models (LLMs). The VBS simulator introduces external events, infusing this data into the LLM system to create specific scenarios. Additionally, images are extracted to (1) enhance the visual aspect of the simulated social computing system, and (2) their embeddings are infused into LLM prompts to generate more life-life digital social interactions.

Results: Our study demonstrates that modeling the impact on public narratives based on simulated scenarios is feasible. The introduction of external events does not compromise the realism of simulated behaviors. Modifying embeddings effectively alters post sentiments, offering an alternative to regenerating posts via modified prompts. Embeddings also facilitate synthesizing images aligned with generated scenarios. Conclusion and Implications: Harnessing LLMs' generative capabilities for simulating diverse social behavior enables predicting the impact of future decisions on public narratives. This approach empowers analysts to experiment with information extraction techniques, develop models, and conduct controlled experiments for training on a large scale in

Chihab Amghane

Affiliation

Refine Results

Social Simulator Madness: Simulating Social Behavior in Dynamic Environments

Mesh-As-A-Service: Automated 3D Modeling Fast as L-AI-ghtning