Tag
This paper studies how LLM-based stance simulation in online discussions is sensitive to counterfactual revisions of conversational context, and proposes an auditing framework comparing text-only and multimodal strategies.