Metadata-Driven Privacy Classification of SQL Server Stored Procedures in Enterprise Web Applications Using Transformer-Based Models

Abstract

Metadata facilitates semantic interoperability, data reutilization, and effective information retrieval across diverse systems. Nonetheless, current research exhibits disjointed methodologies and fails to provide a unified perspective on the tactics employed across many fields. This fragmentation results in a deficiency in comprehending the present state of the art and recognizing cross-domain trends and issues. This systematic study seeks to fill this vacuum by examining metadata integration strategies across many disciplines, including Health and Medicine, Smart Cities and IoT, Data Science, Geosciences, Cultural Heritage, and Library and Information Science. This evaluation adheres to the Kitchenham framework, which includes the preparation, execution, and reporting stages. A total of 81 peer-reviewed articles published from 2014 to 2023 were selected from five major databases, adhering to established inclusion and exclusion criteria and a systematic quality evaluation method. The findings indicate a majority of ontology utilization, succeeded by metadata-driven language, standards, processes, and standardized metadata schemas, alongside a burgeoning tendency towards automation via the implementation of AI-based methodologies. The recognized limitations encompass semantic heterogeneity, insufficient standardization, restricted automation, and usability concerns in existing tools and systems. We offer a thorough synthesis of current methodologies, emphasizing both domain-specific and cross-domain trends, while pinpointing research opportunities to enhance metadata integration via automation, semantic enrichment, and standardized procedures.