From raw museum records to a knowledge graph

The collection, and why it is a fair test

The Museum of Design in Plastics (MoDiP), at the Arts University Bournemouth, is the UK's only accredited museum devoted to plastics in design. It publishes 11,865 object records through the new Museum Data Service under a CC BY 4.0 licence. The records are genuinely rich, with materials, techniques, makers, places, inscriptions and dimensions, and genuinely raw: a good, honest test of what it actually takes to make museum data computable, rather than a cherry-picked clean sample.

The problem, quantified

Across the set there are 476 distinct material strings for what is really a much smaller set of polymers. The same material is recorded under an abbreviation, a full chemical name and one or more trade names, as unrelated free text. A curator or researcher searching for one spelling silently misses the others. These are not edge cases; they are the most common materials in the collection.

What we built, using open standards only

Nothing here is a home-grown schema where a standard already exists. The vocabularies are SKOS; the instance graph is CIDOC-CRM (ISO 21127, the reference ontology for cultural heritage, Linked Art compatible); the materials are aligned by query to the Getty Art & Architecture Thesaurus. A published Spectrum-to-CIDOC-CRM crosswalk, which the sector has lacked in open form, maps each catalogue field to its standard property.

• A materials taxonomy of 137 concepts grounded in polymer science (thermoplastic, thermoset, elastomer, biopolymer), with every abbreviation and trade name folded in as an alternate label, so a search for "polycarbonate" now also finds "PC" and "Lexan". It resolves 99.9% of 35,172 material assertions; the 55 concepts with an exact Getty AAT peer carry a verified link, and the ones without are left unlinked rather than guessed.
• Process and use-domain taxonomies covering 100% of their assertions.
• A 485,000-triple CIDOC-CRM graph over all 11,865 objects, with production events, makers, places, dated time-spans, inscriptions and parsed dimensions.
• A variant graph: 289 object-to-object edges recovered from accession numbers that curators wrote into descriptions ("the same box but a different colourway"), turning relationships readable only by a human into a graph a machine can traverse.

The honesty commitment

The build is a gate, not a claim. SHACL validates at zero violations, and a closed-world vocabulary check confirms every material, process and domain concept referenced by an object is actually defined, with zero dangling references (the check that open-world validation silently passes). What could not be done is written down, not smoothed over: 49 obscure single-occurrence trade names remain unresolved and are listed by name; Getty alignment is partial by design; noisy dates yield only a year. The whole thing regenerates from the committed data with one command and revalidates.

Where this goes

This is one worked example of a pipeline that applies to any of the 250-plus collections already in the Museum Data Service, and to any museum sitting on a Spectrum or CSV export. It is part of Tesseract's Open Ontologies work and shares its method with our heritage aerial-photography and nature-governance graphs: model the domain as a sourced, validated graph so its data can be queried and checked, not just published as a download. If you run a collection, or fund one, and want your catalogue made this queryable, we would like to hear from you.

Independent, self-initiated open research. Source records © the Museum of Design in Plastics / Arts University Bournemouth, published CC BY 4.0 via the Museum Data Service; this transformation released under CC BY 4.0. Endorsed by neither MoDiP nor the Museum Data Service.