Facebook will now screen photos and videos for fake news across 17 countries through its 27 independent third–party partners. It says that it has branched into these two media because the medium of misinformation varies across countries: for example, in the US, misinformation spreads through articles, but in Indonesia, misleading photos are the culprit. Some of the fact–checkers are AFP in Mexico, Colombia, Canada, Pakistan, BOOM in India, Rappler in Philippines. The process of fact–checking Facebook will look at doctored images and videos, and misleading content, or text and audio claims that are provably false. It has built a machine learning tool to identify and flag fake information, which it will forward to independent third-party fact–checkers for review. Fact-checkers will use reverse image searching and analyze metadata to verify the veracity of the visual content. It will use optical character recognition (OCR) to read text on photos and compare that text to headlines articles that have already been flagged by fact-checkers. However, OCR is unable to contextualise flagged visual content. Last week, Facebook announced a machine learning tool 'Rosetta' which could read text from images, and feed the information into a text recognition model “that has been trained on classifiers to understand the context of the text and the image together.” Fact–checkers also consult experts, academics or government agencies to assess the content. Facebook said it that the machine learning model will become more accurate as it gets more ratings from fact-checkers on photos and video. Partnerships to combat fake news In…
