ORES is a web service that provides machine learning as a service for Wikimedia Projects, like Wikipedia and Wikidata. The system is designed to help human editors perform critical wiki-work and to increase their productivity by automating tasks like detecting vandalism and removing edits made in bad faith. ORES is developed by Wikimedia Scoring Platform, a team that specializes in building transparent, auditable, open, and ethical artificial intelligence (AI) to support human decision-making.
ORES is intended to be used as a source of structured information by volunteer tool developers and product developers at the Wikimedia Foundation and Wikimedia Deutschland. Most users access ORES via 3rd party tools like Huggle and Special:RecentChanges on Wikimedia wikis. To access ORES scores, a simple scores API and a reference UI are available.
Scores API
There are three versions of the scoring API that differ slightly in their behavior. Version 3 is the current and recommended version of the API. It supports all current features of ORES (model information, feature injection, and threshold optimization). Version 2 and version 1 are provided for backwards compatibility.
- Version 3 (docs)
- The current recommended
version of the API.
Accessible via /v3/scores/. - Version 2 (docs)
- Preserved for backwards
compatibility.
Accessible via /v2/scores/. - Version 1 (docs)
- Preserved for backwards
compatibility — this version of the API implements ORES
original behavior.
Reference UI
ORES provides a simple user interface for obtaining scores. See user interface page.
Version information
For versions of our dependencies and their versions, see the version page.
Supported wikis
Currently, ORES supports a limited set of wikis; your wiki may be among them. If your wiki is not listed on the table of supported wikis, you can have ORES enabled. See our documentation on mediawiki.org for more information about what models are available and how to get started.
Further information
To read more about the service, see: ORES page at mediawiki.org and our FAQ page.