Research on Methods of Large Language Models in the Field of Sensitive Data Governance

With the rapid advancement of big data and artificial intelligence technologies, the governance of sensitive data has become a critical and pressing issue. Traditional data de-identification approaches exhibit inefficiencies and a lack of flexibility when processing unstructured data. This paper introduces a sensitive data governance method based on Large Language Models, leveraging the natural language processing capabilities of LLMs to construct prompts that enable the identification of fields requiring de-identification and the selection of appropriate de-identification methods. Furthermore, it harnesses the LLMs to automatically generate Python processing code, thus circumventing the limitations of singular and inflexible built-in de-identification functionalities in big data platforms. The objective of this study is to reduce the technical barriers in sensitive data governance and to enhance the flexibility and efficiency of data processing. The paper provides a detailed exposition of the technical implementation of this method, encompassing multiple stages such as data identification, instruction processing, model invocation, result adjudication, and data caching, and discusses its advantages and potential value in practical applications.