Wikidata:Automating descriptions

Thoughts and ideas about generating a description on-the-fly when Wikidata lacks one in the user's language. Commonly called "AutoDesc" after Magnus Manske's tool. For instance, when an Indian city has no description in Spanish, a script or service could generate "ciudad en la India". Of course a human can write a better description, but something is better than nothing.

Use of Wikidata descriptions

The mobile web view and mobile apps show the Wikidata description in search results and on the lead image of an article, but it is missing for most articles in most languages.

Saving descriptions or generating them on the fly

It is better to generate descriptions on-the-fly rather than have a bot that saves descriptions in Wikidata entries because:

• Automatic descriptions will improve over time as
  • editors add more Wikidata properties that AutoDesc uses in its description
  • developers improve the code and linguistic support.
• Most use cases for displaying automatic descriptions will prefer an item's stored description in Wikidata, if it exists. But if a bot has stored a poor description, users will never see a better automatic description, unless complicated logic tries to infer if the stored description came from a bot and AutoDesc can do a better job.
• There's no reason to store generated information. We can cache an automated description for performance yet generate a fresh one.

Tools

• Magnus Manske's AutoDesc.
  • sample long output with Reasonator links
  • sample short text description
  • and fork
• w:en:MediaWiki:Wdsearch-autodesc.js is a user script that displays automated item descriptions in search results on Wikidata.org.
• User:Jitrixis/autoEdit.js, script to automate descriptions.

See also

• Wikidata:Article placeholder input a related proposal to generate articles on the fly ("Reasonator")
• Start of mailing list thread about automatic descriptions
• Meta:Abstract descriptions