The publisher landscape

Who Wikipedia reaches for, by citation rows

Strip a citation record down to its publisher and Wikipedia's reading list appears. Each tile below is one publisher; its area is the number of citation rows in the release-2606 citation spine that name it, its color the medium those citations mostly claim. The three biggest tiles hold a surprise only if you expect scholarship: the most-cited publisher in English Wikipedia is not a university press or a newspaper of record — it is Billboard.

Figure 3. The 300 most-cited publishers of release 2606 as a treemap — tile area is citation rows naming the publisher, tile color its dominant medium — where Billboard's 176 225 citation rows outrank The New York Times (122 937) and Oxford University Press (94 800). Labels appear only where a tile is large enough to hold one with real padding; narrow frames fall back to the publishers' standard short forms (NYT, OUP) so the biggest tiles stay named. Every tile answers to hover, and the treemap is a single tab stop whose arrow keys walk tile by tile. Denominator: the receipted 29.4M-row citation spine of citation occurrences; the 300 publishers shown carry 4 781 700 of those rows, about 16%.

Source: K6-citeworks receipt (A2 publisher landscape) over the citation spine, via the dataset page.

Popular culture is a citation giant

Four of the ten largest tiles are entertainment reference works — Billboard, BBC Sport, AllMusic, and YouTube — and the Guardian and BBC sit just as close to the presses. Oxford (#3), Cambridge (#6), and Routledge (#9) keep university publishing on the board, but the corpus's everyday sourcing muscle lives in music charts, sports wires, and databases that answer editorial questions at scale: when was this single released, what is this player's record, does this album exist.

The color mix says where those questions get answered: of the 4 781 700 rows shown, 52.9% claim the web as their medium and 20.4% news, against 15.2% books and 2.9% journals. Wikipedia is, by citation volume, a web-native encyclopedia with a newspaper habit — the journal tile, so dominant in academic citation, is a sliver here.

How a row counts — and how often it parses

Three conditionings from the receipt govern the numbers. First, rows are citation occurrences, not pages: a publisher cited on 50 pages with two citations each counts 100. Second, names are canonicalized — raw publisher strings merge under one normalized slug, and the tile's label is the most common raw spelling behind that slug, so spelling variants share a tile and the odd name collision is possible. Third, the payload carries each publisher's parse rate: the share of its rows the citation parser read cleanly.

That parse rate draws the campaign's next fight in one glance. Oxford University Press parses at 82.1% and Cambridge at 88.9%, but Billboard — the single largest tile — parses at only 34.3%, and BBC Sport and YouTube sit near 16%. The publishers Wikipedia cites most are template-dense popular-culture sources that our parser reads worst, which is exactly where systematic repair of failed and suspicious citations would pay most.

The full 300-publisher payload, with per-medium mixes and parse rates, lives in the figure's data payload, described on the dataset page.