Synthetic social media creation for social-cybersecurity training

Abstract This paper discusses the creation of realistic, dynamic, and controllable synthetic social media data to support instruction on evaluating social-cybersecurity maneuvers in social media. We propose an agent-based simulation called SynX that takes as input the scenario templates created by Netanomics’ AI-Enabled Scenario Orchestration and Planning (AESOP) tool and outputs an X/Twitter API v1 message corpus by leveraging a large language model (LLM). We conduct an experiment on LLM prompting and evaluate the output of SynX using network metrics and the BEND framework.