Online political inquiry has become a vital channel for governments to address the major concerns of citizens, yet accurately identifying and prioritizing such concerns within large-scale unstructured texts remains a methodological challenge. This paper introduces an integrated framework that leverages large language models (LLMs) to distill core demands, applies UMAP and HDBSCAN for fine-grained topic clustering, and employs Importance–Performance Analysis (IPA) to identify top-priority issues. A regression model is further employed to evaluate the effectiveness of different government response strategies. Evidence from the People’s Daily Online “Leaders’ Message Board” demonstrates that the proposed framework effectively captures diverse citizen expressions and highlights the need for governments to respond promptly and convey positive sentiment in addressing major public concerns.