Facebook extended the fact-checking system to photos and videos posted on the social network. On Thursday (13), the company announced that it is releasing a technology that analyzes the veracity of images for its partners around the world, certified by the International Fact-Checking Network. So far, independent verifiers have focused on articles and written posts.
The new mechanism uses machine learning to facilitate the identification of fake news, based on engagement metrics. The campaign is yet another effort in the fight against fake news promoted by the social network, which currently has 27 marketing partners. fact-checking in 17 countries, including Brazil.
READ: Five things you should not post on Facebook; understand
Facebook expands fight against fake news for photos and videos Photo: Luciana Maline / dnetc
Want to buy cell phones, TV and other discounted products? Meet Compare dnetc
How Image Checking Works
The machine learning model adopted by Facebook uses several signs of engagement, such as feedback in post complaints. These contents are flagged as potentially false by the social network, which forwards them to independent verifiers. Parallel to this, fact-checking partners can also discover photos and videos that look fake on their own and, from there, start the investigation.
To assess the veracity, professionals use techniques such as reverse image search and analysis of photo or video metadata. Typical character recognition (OCR) is also used, a mechanism that extracts text from photos, to compare what is written in the images with headlines from articles already verified.
Example of fake news shared by photo and video on Facebook Photo: Divulgao / Facebook
Facebook also said it was working on new ways to detect manipulated videos or photos, improving the sorting of content that looks fake before they go through the manual review stage. Traditional journalistic clearance procedures, such as surveys with specialists, academics or government agencies, are also employed to ensure the accuracy of the analysis.
According to Facebook, many of the partners they work with already had image verification know-how, but now the system will be expanded to everyone. The model should become even more precise as the evaluations of fact-checking partners increase.
What changes in the news feed
After going through the analysis, a content can be identified in eight ways:
- False;
- Mixed: gathers true and false information;
- True;
- Stira;
- Opinion;
- Ineligible: there is no claim that can be considered true or false;
- Prank generator: sites that allow users to create their own fake news for publication on social networks;
- Not classified: standard status of all publications before evaluation, indicating that nothing should be done about it.
Only the first two classifications false and mixed will incur some type of penalty for publication. Immediately, these contents will have a reduced distribution, being displayed further down in the news feed, so that fewer people will see it. The posts will be accompanied by articles related to the topic, indicated by the fact checkers.
When someone tries to share the publication, they will receive notification of the reports, showing that this is a lie. The notice will also be sent to news that has been shared in the past and has come to be identified as false. Thus, the expectation that this type of post is not as frequent on Facebook.
Most common forms of photos with false information circulating on Facebook Photo: Divulgao / Facebook
Facebook: how to prevent check-in page? Ask questions in the dnetc forum.
Facebook: tips to prevent onlookers from looking at your information