Pacing Opinion Polarization via Graph Reinforcement Learning

This arXiv feed collects multiple 2026 papers about social/graph methods and LLM-agent ecosystems with direct relevance to information operations and platform safety. Key contributions: PACIFIER (graph reinforcement learning) presents sequential network-intervention policies for pacing opinion polarization and budget-aware or topology-changing moderation; several empirical studies (Chirper.ai, Moltbook, and an LLM-agent contagion model) show LLM-driven platforms and agent networks can rapidly concentrate attention, amplify news, and exhibit polarization and toxic behaviors; Clone-Robust Weight