Artists' Books

A linked data publication pipeline that transforms a curated Zotero library into a knowledge graph and static website. The collection documents artists' books held at the Joseph C. Sloane Art Library (UNC Chapel Hill), along with citation data tracking how those books appear in reference literature.

Pipeline

The pipeline runs as two parallel tracks that meet at the graph stage: the artists' books (the 1,341-book collection the site is built from) and the reference works that cite them (the 157-item Reference resources collection, plus the "Cited:" notes connecting the two). Both are surfaced on the website and cross-linked — a book lists the reference works citing it, a reference work lists the books it cites. Only reference works that actually cite an in-collection book get a page (43 of the 157 today).

See docs/PIPELINE.md for a useful diagram, and docs/README.md for an index of the supplementary documentation.

The stages, and where each is documented in depth:

1. Zotero → CSV/notes. dedup.py collapses the three Zotero libraries into the canonical book (~7.9k) and reference-work lists; notes_export.sh emits the "Cited:" note paragraphs; and freeze_citations.py reconciles those paragraphs to reference works and freezes the citation edges into sources/citations.ttl. Both fuzzy steps are gated by a hand-owned decisions overlay — a person's same/no/unsure calls in sources/artists-books-dedup-decisions.csv (per lib-1 key) and yes/no/unsure calls in sources/citations-decisions.csv (per reference-work × book pair). These are read back but never rewritten by make, so a re-run never clobbers them; each step also emits a generated *-review.csv surfacing the uncertain matches to curate. → sources/zotero/README.md (database, notes, citation data model), sources/README.md (canonical dedup + frozen citations).
2. Library catalogs → MARCXML. marc_harvest.py (YAZ) harvests full MARC records over Z39.50/SRU — books from UNC's catalog, reference works from a chain of nine catalogs by ISBN then title+author — each stamped with a synthetic 999 $a <itemKey> to join back to its Zotero item. → sources/marc/README.md (harvest), docs/MARC-RECORDS.md (field-level analysis).
3. CSV + MARCXML → RDF graph. Two SPARQL-Anything CONSTRUCT queries (queries/artists-books.rq, queries/reference-works.rq) emit BIBFRAME plus a custom ab: namespace into graph/*.ttl. The citation edges are not constructed here — they are the frozen sources/citations.ttl (issue #82): 5,022 citations link 3,896 books to 54 reference works. Two parallel local SKOS vocabularies are mined from the book MARC and curated by hand, each gated by its own decisions.csv overlay (include? / concept / category, per heading cluster; hand-edited, never rewritten by make): construction techniques, materials, binding/format types and printing methods (how a book is made) in sources/construction/decisions.csv → sources/construction-methods.ttl, and topical/geographic subjects (what it is about) in sources/subjects/decisions.csv → sources/subject-terms.ttl. Their build-scheme.rq queries emit the schemes, and artists-books.rq attaches an ab:constructedUsing / ab:hasSubject link from each book to the concepts its headings map to. → CLAUDE.md (query architecture, URI minting), sources/construction/README.md and sources/subjects/README.md (the mine → curate → build vocabulary pipeline), docs/QUERY-PERFORMANCE.md.
4. RDF graph → website. Apache Fuseki loads the two constructed graphs, the frozen citations, and the construction-methods.ttl and subject-terms.ttl SKOS schemes, and serves a local SPARQL endpoint; Snowman runs the SELECT queries in web/queries/ against it and renders the Go templates in web/templates/ into web/site/ — including, on each book page, the construction-methods and subject sections resolved from the schemes. → CLAUDE.md (views, templates, Snowman gotchas).

Building

make all       # build graph/*.ttl, then web/site/index.html
make serve     # build if needed, then serve at http://127.0.0.1:8080

All tools are fetched (and, for YAZ, built from source) on first use — no third-party binaries need a system-wide install. The only system prerequisites are a JVM, sqlite3, make, and a C toolchain (to compile the vendored YAZ); GitHub Codespaces gets these via .devcontainer/devcontainer.json. See CLAUDE.md for the full target list and build details.

Regenerating the inputs under sources/ additionally needs GNU Make ≥ 4.3, for grouped targets. macOS ships GNU Make 3.81, so there use gmake (brew install make) — make -C sources … stops with an explanatory error rather than misbuilding. The make all / make serve build above has no such requirement.

Tool	Version	Purpose
Apache Jena	6.1.0	RDF validation, RDFS reasoning, graph diffing (riot, arq, shacl)
Apache Fuseki	6.1.0	In-process SPARQL endpoint server
SPARQL-Anything	1.1.0	CSV/MARCXML-to-RDF transformation via SPARQL CONSTRUCT
Snowman	0.8.0	SPARQL-driven static site generator
YAZ	5.37.3	yaz-client/yaz-marcdump — Z39.50/SRU MARC harvest; built from source

Vocabulary & data model

The graphs emit BIBFRAME with a custom ab: namespace layered on top (ab:ArtistsBook, ab:ReferenceWork, ab:Citation, ab:cites/ab:citedBy, ab:constructedUsing, ab:hasSubject, creator-role properties). The concepts that ab:constructedUsing and ab:hasSubject point at live in companion SKOS schemes, sources/construction-methods.ttl and sources/subject-terms.ttl — mined from the book MARC and curated by hand (see sources/construction/README.md and sources/subjects/README.md). docs/vocab.ttl defines the vocabulary and docs/description.ttl is a hand-written worked example (Ed Ruscha's Twentysix Gasoline Stations); make validate runs Jena's validator over both. Some terms are defined but not yet emitted, and the citation terms still carry a legacy ex: prefix pending normalization — see docs/README.md and CLAUDE.md for current status.

The Zotero library encodes the citation data three ways — tags, "Cited:" notes (~7,649), and dc:relation links (~2,366); the notes are the form the pipeline extracts. See sources/zotero/README.md → Citation data for the full model.

Relevant resources

• Opening Artists' Books vocabulary
• Book Arts Research Database
• Artists' books with authors in Wikidata
• Joseph C. Sloane Art Library Collection of Artists' Books and Zines
• Artists' Books in the Center for Book Arts collection