Linked data has been promised to libraries for more than fifteen years. The pitch is simple. Instead of describing a book with strings of text, you describe it with links to shared, machine-readable identifiers for people, places, subjects and works. Machines can then connect your records to everyone else's.
The reality has been slower. Large national libraries have invested heavily. Most small catalogs still run on MARC records and a vendor system that will not change soon. This post explains the main pieces in plain terms and suggests first steps that make sense for a small catalog today.
What linked data means for a catalog
Linked data is a way of publishing data on the web. It rests on a few ideas. Things get identifiers in the form of URIs. Those URIs resolve to useful information. Statements about things are expressed as small triples: a subject, a property and a value. And values point to other URIs whenever possible, not just to text.
In a traditional record, an author heading is a string. Two catalogs can spell the same person differently, and a machine has to guess whether they match. In a linked record, the author is a URI from an authority file. If two catalogs use the same URI, the match is certain. If they use different URIs that are linked to each other, the match is still reliable.
That is the real value. Linked data is not mainly about new formats. It is about replacing ambiguous strings with stable identifiers that many systems share.
The main pieces: BIBFRAME, Wikidata and authority files
Three things come up in almost every linked data conversation. They do different jobs, and it helps to keep them apart.
- BIBFRAME is a data model developed under the leadership of the Library of Congress as a long-term successor to MARC. It describes bibliographic resources as linked entities, mainly works, instances and items. BIBFRAME 2.0 has been the working version since 2016, and conversion tools from MARC are freely available.
- Wikidata is a free, collaboratively edited knowledge base run by the Wikimedia Foundation since 2012. It holds items for people, places, organizations and concepts, and it links many of them to library authority identifiers. Anyone can edit it, which is both its strength and its risk.
- Authority files are the controlled vocabularies libraries already use. The Library of Congress publishes its name and subject authorities as linked data at id.loc.gov. VIAF, hosted by OCLC, clusters name authorities from many national libraries. FAST offers a simplified faceted subject vocabulary derived from LCSH.
For a small catalog, authority files are the most useful starting point by far. You already use their headings. Adding their URIs to your records is a modest change with a direct payoff.
Comparing your options
The table below summarizes what each piece offers a small library and what it costs in effort.
| Option | What you gain | Effort | Main risk |
|---|---|---|---|
| Authority URIs in MARC ($0 and $1 subfields) | Reliable matching of names and subjects across systems | Low to moderate, often batch work | Your system may not display or preserve the subfields |
| Wikidata items for local entities | Visibility for local people and places that national files miss | Moderate, ongoing editing | Open editing means data can change without notice |
| Full BIBFRAME conversion | A catalog ready for linked data exchange | High, needs tools and staff time | Few small-library systems use BIBFRAME natively yet |
| Linked data export only | Your records reusable by aggregators and researchers | Moderate, a script or plugin | Export can drift out of sync with the catalog |
Most small catalogs should start in the first row and move down only when a concrete need appears.
Realistic first steps for a small catalog
You do not need a new system to begin. You need clean headings, a plan, and patience.
- Clean your headings first. Linked data magnifies messy data. Fix obvious variant forms of names and subjects before you attach identifiers. Our post on metadata quality metrics covers ways to measure where the problems are.
- Add authority URIs where they already match. Many systems and batch tools can look up headings against id.loc.gov or VIAF and add the URI in a $0 or $1 subfield. Start with personal names, where the match rate is usually highest.
- Check that the URIs survive. Export a sample, re-import it, and confirm nothing was stripped. Some older systems drop unfamiliar subfields silently.
- Describe local entities in Wikidata. Local authors, photographers, organizations and buildings are often missing from national files. A carefully sourced Wikidata item gives them a shared identifier others can reuse.
- Expose what you have. If your system can publish records through OAI-PMH or an API, include the URIs there. Aggregators can use them even if your own interface does not.
None of these steps commits you to BIBFRAME. All of them make a later move easier.
Plan the work in small batches. A few hundred records at a time is enough to find problems without creating a backlog of errors. Keep a simple log of what was changed, by which tool, and on what date. If a vendor update or a migration later strips the identifiers, that log tells you exactly what to restore. It also gives you something concrete to report: how many headings now carry a URI, and how many still need review. Those numbers make the case for the next phase far better than a general promise about the semantic web.

Lessons from the harvesting era
The current interest in linked data echoes an earlier wave of work on shared metadata. In the 2000s, the MetaScholar Initiative, a digital library program run from Emory University's Woodruff Library, built tools for combining records from many institutions. Its Metadata Migrator helped museums, archives and small libraries map local fields to Dublin Core and publish them for OAI-PMH harvesting. The MetaCombine project then explored how to cluster and enhance records gathered from mixed sources.
That work met a familiar problem. Harvested records were full of inconsistent names, dates and subjects, and aggregators spent much of their effort reconciling strings. Shared identifiers address that problem directly. A heading that carries a URI does not need to be guessed at downstream.
The other lesson is about scale. The projects that lasted were the ones small institutions could actually run. A tool that maps a spreadsheet to a standard format beats an elegant model nobody can maintain. The same applies now. A catalog with good authority URIs and a working export is more useful to the wider web than a half-finished conversion. Our post on Dublin Core at thirty makes a similar argument about simple metadata.
What to avoid
A few mistakes come up often. Do not mint your own URIs for people and subjects that already exist in national files; link to the existing ones. Do not import Wikidata statements into your catalog without review, because open data can be wrong. Do not treat a pilot conversion as a production system unless you have a plan to keep it in sync.
Linked data rewards steady, incremental work. Each identifier you add is useful on its own, today, whether or not the full vision ever arrives.
