pipette
ENEnglish

WPTokens: A Dataset for Studying the Regulation of Wikipedia

L\'eo Joubert

Preprint

En palabras de los autores

Studying how Wikipedia is regulated requires knowing who adds, rewrites, moves and deletes which piece of content, when and where. WPTokens is a dataset that answers these questions over the full revision history of the French-language Wikipedia. It relies on a positional diff over two layers: a text layer made of sentences, and a node layer made of the structural elements of wikitext (templates, links, references, headings, tables, lists, etc.). Each piece of content keeps a stable identifier for its whole life, and every change to it is recorded as an addition, an update, a move, a move with update, or a deletion. Built from the April 2026 dump, the dataset covers 2.75 million articles, 153.6 million revisions, 342.6 million tokens and 573.7 million token events, stored in an indexed SQLite database. A companion R package turns these records into temporal contributor--token multigraphs, completed by a layer of revert ties between contributors. We describe the tokenization, the diff algorithm, the data model and the graph construction, and discuss the limitations of the approach.

Resultado principalLimitación que admiten los autores

Apareció: miércoles, 7 de octubre. arXiv. Preprint, todavía sin revisión por pares.