Bridging library metadata to the semantic web sounds simple — until you're staring at a fuzzy title match and wondering whether "Emma" the Wikidata item is actually "Emma" the Jane Austen novel. This talk walks through a working pipeline that reconciles MARC records and spreadsheets against Wikidata before generating BIBFRAME RDF — deliberately in that order.
MARC's fixed fields, controlled identifiers, and $0/$1 authority links are a more reliable reconciliation substrate than post-conversion RDF would be: decades of cataloging discipline mean an ISBN or an LC name-authority code can be trusted at face value, in a way a freshly-generated triple can't yet be. Reconciling at this stage — while those signals are still intact — lets Wikidata's enormous, freely-reusable graph of works, people, and places get folded directly into the RDF as it's built, rather than bolted on after. MARC turns out to be less a legacy format to escape and more infrastructure in service of the linked data it produces.
The core mechanism is a tiered confidence system: identifiers auto-accept, fuzzy titles get a second look only when independently confirmed by author identity, and everything else is flagged rather than guessed at. I'll also cover verifying the BIBFRAME modeling itself against real production records — which caught errors theory alone missed.
