Data Packages: a Nucleus for Enterprise Knowledge Graph Evolution

Knowledge Graphs (KGs) are increasingly used to power users and business applications. Yet, KG construction is often an ad hoc activity, resulting in management and scalability challenges, especially in large enterprises. Key challenges are transparency, provenance, and lifecycle management. Existing methods lack standardized approaches for encapsulating data or managing process dependencies that make it inordinately difficult for non-expert users to extend KGs without a deep understanding of RDF technologies. Furthermore, advanced GenAI and Machine Learning methods often produce new artifacts from KGs -- such as embeddings, clusters, and models -- which can be represented as enhancements to the KG but are often stored in a governance vacuum, disconnected from the KG's lifecycle. To address these limitations, we introduce a novel RDF-based data package management framework where knowledge is organized into modular, self-contained and self-described units called \emph{data packages}. Each data package encapsulates RDF content, rich metadata for lineage and versioning, SHACL shapes to which the data content adheres, and comprehensive documentation and vocabulary graphs. This approach facilitates precise provenance tracking, robust structural validation, and modular graph re-use. By introducing abstraction layers inspired by the Croissant standard, we provide a structure interface that enables domain experts to safely extend and reuse graph content. This research focuses on achieving reproducibility, transparency, composability, and scalability for enterprise-level knowledge graph operations. Ultimately, our approach fosters broader adoption and drives rapid, domain-driven evolution of KGs in enterprise settings.