LLM-Augmented Digital Twin for Policy Evaluation in Short-Video Platforms 事件

LLM-Augmented Digital Twin for Policy Evaluation in Short-Video Platforms arXiv:2603.11333v2 Announce Type: replace Abstract: Short-video platforms are closed-loop, human-in-the-loop ecosystems where platform policy, creator incentives, and user behavior co-evolve. This feedback structure makes counterfactual policy evaluation difficult in production, especially for long-horizon and distributional outcomes. The challenge is amplified as platforms deploy AI tools that change what content enters