Research on Automatic Government Data Cataloging System Based on Data Architecture and Local Large Language Model: A Case Study of Suining City

With the in-depth advancement of city-wide digital transformation, government departments generate massive data that is challenging to inventory and manage effectively. Traditional cataloging systems rely heavily on manual input—a time-consuming, error-prone approach that cannot adapt to dynamic data environments. Additionally, data privacy and security constraints often preclude the use of cloud-based solutions, and there is a dearth of specialized tools to automate the cataloging process. To address these issues, this paper proposes an automatic data cataloging method based on a "One Body, Two Wings" Data Architecture (DA) and a local large language model (L³M). In this concept, the "One Body" refers to a unified data foundation, while the "Two Wings" denote intelligent governance and secure sharing. The proposed system automatically obtains metadata and integrates knowledge graphs with semantic classification to achieve automatic directory generation, label optimization, and dynamic catalog updates. Experimental results on three government systems in Suining City (urban management, citizen card, and cloud imaging) show that the classification accuracy reaches approximately 85%, significantly reducing the manual workload and labor costs. This work demonstrates the practical potential of deploying local large language models to obtain secure, efficient, and intelligent government data governance.