Wikidata:Third Birthday/Presents/ORES
Happy Wikidata's birthday! Amir and I have a gift. We've successfully deployed a machine learning model for detecting vandalism in Wikidata. If you want to see what the API does, you can try scoring an edit like this:
http://ores.wmflabs.org/scores/wikidatawiki/reverted/247205553/
But that's a machine readable interface, so I recommend installing one of our simple gadgets that takes advantage of ORES. In order to enable the gadgets, you'll need to edit one of your "user JS pages". I recommend editing the global.js on meta since it affects all wikis and this script will work anywhere that ORES is available.
Step 1. Go to:
https://meta.wikimedia.org/wiki/Special:MyPage/global.js
Step 2. Copy paste these lines onto the page and hit save:
// [[File:User:He7d3r/Tools/ScoredRevisions.js]] (workaround for [[phab:T35355]])
mw.loader.load('//meta.wikimedia.org/w/index.php?title=User:He7d3r/Tools/ScoredRevisions.js&action=raw&ctype=text/javascript');
Step 3: Go to Special:RecentChanges on Wikidata:
https://www.wikidata.org/wiki/Special:RecentChanges Step 4: Hard refresh your browser to pull in the new javascript
You should see something that looks like this:
The highlighted edits are scored as likely to be vandalism by ORES. As the color turns from yellow to red, ORES's confidence that the edit is vandalism increases.
Right now, the fitness isn't great. There's some vandalism we miss and some non-vandalism that we flag. We'll be working on the prediction strategy to improve accuracy over the next couple of months.