Wikidata - Publishing Knowledge Graph + SPARQL

It is the connective tissue of book metadata, filed under Publishing and scoring 9 out of 10 on Datadory's quality rubric - inside the top tier of the sixteen-record publishing slice we catalog. The knowledge base held 122,983,238 entities when the August 2026 research pass read its counter, and book publishing lives inside it as structured statements on item entities: ISBN-13 (P212) and ISBN-10 (P957) pin the edition, title (P1476) arrives language-tagged, publisher (P123) and author (P50) resolve to their own nodes, publication date (P577) grades to year-month-day precision, and P291, P31 and P921 carry place of publication, work class and subject.

Two things make it analytic rather than merely encyclopedic. First, the links are the product: publisher-book-author triples across languages turn a pile of titles into a network you can walk, which is why publishers' houses, imprint relationships and multilingual editions of one work come out as queries rather than reconciliation projects. Second, the verification was live: the research pass confirmed publisher, title, ISBN-13 and publication-date claims on a real book item, so the dictionary below describes observed statements, not documentation aspiration. Get a sample of this dataset to inspect real rows before committing pipeline time.

One row per book-bearing item, wrapped in the graph's own identifiers:

# one edition row - delivered grain is one row per book-bearing item entity
item                    : Q100708395
title (P1476)           : Silver Economies, Monetisation and Society in Scandinavia, AD 800-1100
isbn13 (P212)           : 978-87-7934-585-0
publication_date (P577) : 2011-01-01

# a second edition shows the same shape carrying another language market
item                    : Q100708549
title (P1476)           : Dictionnaire historique des Academiciens de Lyon
isbn13 (P212)           : 978-2-9559433-0-4
publication_date (P577) : 2017-03-01

# every row resolves outward through typed statements, not loose text
publisher (P123)        : -> publishing-house item
author (P50)            : -> author item
instance_of (P31)       : -> book / literary work class

# graph-wide scale accompanies the rows
entities.total          : 122983238        # counter read August 2026

Read what the pinned row proves. The item Q-number anchors every downstream join and never gets recycled; the ISBN-13 ties the row back to the trade catalogs that speak barcode; the language-tagged title means a Portuguese edition and an English edition of one work stay distinct rows that still point at shared author and publisher nodes. The two count lines matter less than the arrow lines beneath them - P123, P50 and P31 arrive as item links, so attribution is a graph fact with its own node, ranks and references in the claims container rather than a free-text afterthought. Graph-wide scale - 122,983,238 entities at the August 2026 reading - travels beside the rows it summarizes. Example rows are illustrative of the record shape; your sample ships pinned to real records cut to your filters.

Eleven verified fields define the item record, every definition checked against live output during the August 2026 research pass rather than inferred from property names. They group cleanly: identity (P212, P957, P1476), people and firms (P123, P50), time and place (P577, P291), classification and topic (P31, P921), and the multilingual apparatus (labels / descriptions / aliases) with the claims container that keeps ranks and references attached to each assertion.

The full dictionary follows in the table below. Note the deliberate asymmetry with ordinary bibliographic feeds: half the dictionary is pointers. Publisher and author are nodes you traverse, which is precisely what makes imprint networks and multi-market edition maps computable instead of hand-built.

The extensions above reshape the same graph rather than invent new facts, and they confirm against your named scope when the sample is cut:

Geography - global and multilingual, with one honest structural note: coverage density varies strongly by language community, because statements accumulate where contributor communities are active. A title hot in one market may have thinner statements than its bestseller cousin elsewhere - measurable per scope, and worth screening at sampling rather than assuming.

Temporal - works from early printing to current titles, and the graph keeps moving: editors and bots add statements continuously, so the corpus is a live fabric rather than a frozen snapshot. That combination serves a printing-history study and a this-quarter new-edition watch from the same dictionary.

Granularity - entity-level throughout: book editions, works, authors and publishers, connected by typed statement properties. Rollups come naturally along any edge - per publisher, per author, per class, per language - so record-level cataloging and market-share arithmetic share one spine. Property presence is uneven by design across real-world books; fill rates for P123, P50 and P212 within your named scope are a number we measure together at sampling instead of a promise made blind here.

API, files, or your warehouse. Daily, weekly, or hourly.

You pick the channel and the cadence; endpoint etiquette, pagination walks and change detection stay our problem. Rows arrive normalized to the field dictionary above and keyed on the item Q-number with ISBN-13 alongside, so consecutive deliveries diff cleanly into new-edition arrivals and statement corrections instead of piling up as undifferentiated snapshots.

Hourly suits teams watching new statements land on titles they track. Daily suits catalog enrichment and deduplication jobs that run between warehouse loads. Weekly suits network analysis, where publisher graphs move slowly enough that snapshots lose nothing. Whichever you pick, definitions travel unchanged and schema shifts are flagged rather than discovered. A sample cut to your named publishers, languages, subjects or date windows comes first either way.

Five uses dominate, and all five lean on the same trick - treating publishing facts as a graph instead of a spreadsheet:

Fit tracks how much of your question lives in the relationships between books, people and firms:

Publishing's shelf splits by job, and this record owns the linkage layer. OpenAlex is the breadth play - roughly 322 million works including 5.9 million books - when citation volume matters more than cross-market identity. Open Library Search & Works/Editions API answers which editions of a book exist inside a lending catalog, while this one answers how that book, its author and its publisher hang together across every language. The International ISBN Agency registers supply the prefix arithmetic - who may mint which ranges - but carry no per-title statements; Google Books and Goodreads hold the retail-facing descriptions and ratings. Our twin record, the Wikidata SPARQL Query Service, sees the same graph from the query-seat angle - the trade-offs are worked through in the dedicated comparison.

If the question is what connects these books, authors and publishers across markets, this is the record.

Because the raw artifact underneath is a living encyclopedia edited by humans and bots, and every interesting question wants thousands of editions, many publishers and several languages in one frame. Datadory hands over the graph itself: the eleven-field dictionary kept verified and current as editing practice evolves, deliveries keyed on the item Q-number with ISBN-13 alongside so consecutive pulls diff into clean timelines, label and alias tables resolved into their own columns, and the whole thing landed on the cadence your pipeline runs - with sample rows in hand before any commitment. Browse the rest of the slice on the publishing data hub, start from the best publishing datasets ranking, or see how the slice hangs together in the publishing data guide.