Earlier, Textgain developed software to decipher hate messages on Twitter. However, terrorists or supporters of terrorist movements adapted their methods, posting encripted messages or communicating via video footage, which computers cannot read.
Researchers have found an answer to this. Using so-called deep-learning techniques, thousands of pictures of IS supporters are being compared to those of other users to learn about the differences.
"Our success rate has now risen to 85 to 90 percent. But give us a million photos more and this percentage can still climb", explains Guy De Pauw. The computer can make the difference simply because IS supporters share different things on the web - such as weapons, explosions, jihadist fighters and flags - compared to citizens - who would post selfies, pets, film heroes, football stars and oven dishes.
The new software can also be used to counter jihadist propaganda online.