Do you want to know how Facebook wants to detect manipulated pictures and videos?
Text is easily checked by algorithms, but what about pictures and videos? Facebook is facing this widespread challenge to identify fake news spread via graphic content.
How is that going to work?
Similiarly to articles there will be a specific algorithm checking graphic content. It uses different automated methods like a online image search to see if that image had already been used elsewhere.
Suspicious pictures and videos get forwarded to extern fact checkers to investigate further. These fact checkers are specialized on the examination of graphic content, e.g. using meta data to check for publication dates and places. They’re being supported by experts, scientists and government agencies. The more evaluations are being fed into the system by fact checkers, the better the systems learns to detect future fake content.
The algorithm uses for example an OCR mechanism. With that text in graphics is being recognized and cross referenced with text from other articles to check for its validity. Furthermore, Facebook is working on techniques to detect rigged pictures even before they are being sent to fact checkers.
How does Facebook categorize potential fakes and hoaxes?
Usually fakes and hoaxes are divided into three categories:
Facebook is working together with indepentent fact checkers to further refine their algorithms. But it’s still a long way to have Fake News vanish automatically. And there is also the possibility that reports on Fake News themselves are wrongly deemed fake since they have to use the forged pictures themselves.
To improve the system as much as possible Facebook is looking for new partners and further refining their algorithm which is in use since 13th September 2018.