Advancing AI-Assisted Policy-Text Analysis: Revealing the Drive–Adaptation Model of Central–Local Governance

Abstract

This study integrates fine-tuned large language models (LLMs) with Query-by-Committee (QBC) active learning to analyze 1,863 Chinese policy documents on new energy vehicles (2009–2024). The approach demonstrates how fine-tuning and active learning enable accurate, scalable, and transparent policy-text analysis across five analytical dimensions. The results reveal a distinctive pattern of central–local coordination—the Drive–Adaptation model—where the central government provides strategic direction while local governments adapt implementation to regional conditions. The study contributes a replicable AI-based framework for planning research and clarifies a planning-relevant governance logic of adaptive coordination in emerging innovation-driven industries.