Invited Talk: AI-Assisted Metadata Enrichment in Collaborative Knowledge Graphs

Collaborative knowledge graphs provide an ideal environment for applying Artificial Intelligence to metadata enhancement, provided that AI-generated knowledge remains transparent, traceable, and aligned with established cataloguing principles. This presentation introduces two complementary AI-assisted metadata enrichment approaches developed within the Share Family, a collaborative, library-driven initiative built on BIBFRAME and supported by a shared entity management infrastructure. The first approach enriches entity descriptions by identifying incomplete metadata and using Large Language Models to infer missing information from existing textual evidence. AI-generated statements are explicitly identified through provenance metadata, ensuring that machine-generated assertions remain distinguishable from human-created cataloguing. The second approach focuses on identifying contributor roles that are often expressed only as free-text responsibility statements in legacy bibliographic records. By combining rule-based analysis with AI-assisted interpretation, the system transforms implicit textual information into structured semantic relationships that improve discoverability and interoperability while remaining consistent with established metadata standards. Together, these approaches demonstrate that AI can enhance, rather than replace, cataloguing expertise. Instead of generating knowledge autonomously, AI operates within a metadata framework that documents, contextualizes, and makes its outputs accountable. Metadata therefore become the mechanism through which AI remains explainable, auditable, and interoperable, enabling the transformation of implicit bibliographic knowledge into explicit, reusable semantic information while preserving the human values of transparency, provenance, and community-governed metadata creation.