fbpx
The Stack Archive

Wikipedia creates AI system to filter out bad edits

Wed 2 Dec 2015

X-ray specs

Free online encyclopaedia Wikipedia has developed a new artificial intelligence (AI) system aimed at improving the quality of its entries and detect mistakes and damaging edits made to its articles.

The technology, named the Objective Revision Evaluation Service (ORES), works like “a pair of X-ray specs,” according to the Wikimedia blog. It explains that the system is able to highlight incorrect edits, allowing editors to filter them out from the “torrent” of new amends and scrutinise their credibility.

Screen Shot 2015-12-02 at 11.22.05Around half a million edits are made on the Wikipedia platform every day – a heavy workload for human editors. As an added pressure, the MIT Technology Review reported that the foundation has lost 40% of its active contributors over the last eight years, limiting it to around 30,000. It is hoped that the new ORES service will help to compensate for this shortfall and growing amount of work that’s required.

The AI system, developed by Wikimedia Foundation’s senior research scientist Aaron Halfaker, functions by analysing quality assessments made by humans (Wikipedians) to approve and disregard edits. The software is able to apply this machine learning to identify patterns and make decisions on its own.

As ORES detects an apparent mistake, a human editor is alerted to the issue and can correct it if needed. Equally, editors who have made an error will be informed in order to learn from their mistake.

For now, ORES will be used to monitor new edits made in English, Farsi, Portuguese and Turkish, with other language versions rolling out soon.

“Our hope is that ORES will enable critical advancements in how we do quality control—changes that will both make quality control work more efficient and make Wikipedia a more welcoming place for new editors,” said Halfaker.

The entire service and process is open – with Wikipedia making revision scoring transparent and auditable by publishing the source code, performance statistics and project documentation publicly under open licenses.

Tags:

AI news research
Send us a correction about this article Send us a news tip