AI-Powered Financial Insights: Using Large Language Models to Improve Government Decision-Making and Policy Execution

Abstract

Given the complexity of fiscal data types and the lengthy policy execution chain, this study constructs an application framework for language models supporting government decision-making. It systematically investigates task modules including decision-making question-answering identification, expenditure forecasting modeling, executive summary extraction, semantic matching, and conflict reasoning. The framework elucidates model architecture design methodologies and semantic fusion mechanisms, while introducing response capability simulation testing and performance evaluation systems. Using heterogeneous fiscal corpora and multi-task experimental data, demonstrates that the model exhibits strong performance in accuracy, generative consistency, and generalization capabilities, supporting intelligent applications across diverse fiscal scenarios.