Tools for truth: Spotting fact and fake news on social media in poll season

Morphed images, mischievously edited videos, and false messages could be flooding your timeline this election season. Here is how you can be your own fact-checker on Facebook, Twitter and WhatsApp

A viral video on Twitter gave the impression that Delhi Chief Minister Arvind Kejriwal was drunk. Kejriwal was sober — the video had been edited mischievously. A TV news channel showed satellite images to prove that the Indian Air Force (IAF) had demolished a terrorist group’s camp in Balakot, Pakistan. The images, picked up from Twitter, were old and unrelated to IAF’s operation. A 2017 image of Congress leader Priyanka Gandhi was photo-shopped to show her wearing a Christian cross and it went viral on Twitter. 

Voters, beware! Social media is being used for spreading fake news and misinformation during Lok Sabha elections 2019.

"Most misinformation in India is in the form of images and videos — with the attached text blurb most often the source of the misinformation — which are shared overwhelmingly over WhatsApp on mobile phones," says Pratik Sinha, co-founder of fact-checking website Alt News.

Facebook, Twitter, and WhatsApp do tackle misinformation, but they cannot check each post, tweet or message. Fact-checking this election season will fall to readers. "It is unlikely that people will sit down and fact-check every piece of information or WhatsApp forward received by them along with their day jobs. For it to be something everyday folk can do actively, fact-checking will have to be a click or two away, needing as little an investment of time as possible," says Sinha.

Image, videos and reality

Fact-checking need not be cumbersome, though. A simple Google reverse image search can establish if a gruesome forwarded photo of a riot victim is recent or has been circulating online for a few years, and whether it is from India or elsewhere.

Alt News used Google reverse image search — not specifically built to track fake news — to debunk the photo-shopped image of Priyanka Gandhi wearing a cross.

Videos, however, are harder to fact-check because of technology. "Deepfake" videos made using artificial intelligence software, such as FakeApp, would appear very real.

However, deepfake videos for misinformation don't appear to be the norm in India. Instead, videos on the internet are re-purposed to change their original context and they are then used to spread misinformation.

An example can be Pakistani social media handles on February 27, 2019, sharing a video claiming to show a captured Indian Air Force (IAF) pilot. This came after Pakistan claimed it had shot down two Indian aircraft inside its airspace and arrested an IAF pilot. An Alt News analysis of the video found it dated back to February 19 — a day before the Aero India Show in Bengaluru when two Surya Kiran copters had crashed during rehearsal, killing one pilot and injuring two others. The video shared on Pakistani social media was actually of a civilian in Bengaluru comforting one of the injured pilots, Vijay Shelke, who had parachuted out of his aircraft.

You could reverse image search a screen-grab of a video, or use Amnesty International's YouTube DataViewer, which provides thumbnails of a YouTube video along with the option of using Google reverse image search on them. 

The InVID Verification Plugin is a more comprehensive toolkit. Here's a list of things you can do with it:

1) Get contextual information on Facebook and YouTube videos

2) Use reverse image search on stills from a video through Google, Baidu or Yandex search engines

3) Fragment videos from platforms such as Facebook, YouTube, Twitter, Instagram, and Daily Motion into keyframes, which can then be put under a magnifying lens

4) Use filters in Twitter search for keywords and hashtags

A video was shared on Facebook in December 2018 apparently showing Muslims burning down Hindu houses in a village in Rajasthan soon after the Congress came to power in the state. Alt News used InVID to establish that it was in truth the footage of an incident in Bangladesh dating back to 2013.

Watch how the December 2018 Facebook video was debunked



 

InVid isn’t perfect though: It didn’t work to find out the original context of the slowed-down video of Kejriwal.

Verifying information or news as text is simpler. Alt News co-founder Sinha says that the thumb rule is to question any information that is provocative in nature. "In the case of WhatsApp rumours about child-kidnapping gangs, which led to lynchings in the past, the intention was to evoke fear among people — that your children might be in danger, they could be kidnapped and trafficked," he explains.

Institutionally, WhatsApp, Facebook and Twitter use varied approaches to fake news and misinformation.

WhatsApp’s way 'forward'

WhatsApp's end-to-end encryption prevents it from moderating messages. "The company cannot access the content of any message sent on its platform and does not actively take down content," says a WhatsApp spokesperson.

However, WhatsApp detects suspicious behaviour by identifying messaging patterns. "We rely on advanced machine learning capabilities that evaluate the behaviour of an account, without regard for the content itself. WhatsApp was built for personal and small-group use and we have always taken a strong stand against spam of any kind. Our detection technology works around the clock to spot accounts engaging in abnormal behaviour so that they cannot be used to spread spam or misinformation. We also rely on user reports to take action against accounts sending automated unwanted messages," says the spokesperson.

Facebook-owned WhatsApp has said that it removes more than 2 million accounts per month globally for “bulk” or automated behaviour. A WhatsApp white paper in February said that more than 75 per cent of those accounts were banned even without a recent user complaint against them. Roughly 20 per cent accounts were banned at the time of registration itself.

According to WhatsApp, an account that was registered five minutes before attempting to send 100 messages in 15 seconds is banned automatically. An account that tries to quickly create dozens of groups or add thousands of users to a series of existing groups is banned, too. "We have also engaged with numerous political parties to ensure they are aware of our terms of service and that attempting to send bulk messages on WhatsApp is not allowed," says the spokesperson.

WhatsApp users can reach out to its support team directly from the app under 'Settings'. If the support team is unable to help, they can contact WhatsApp’s grievance officer, Komal Lahiri.

"Users looking for help from fact-checking organisations can message Boom Live, Alt News, WebQoof, Factly, Newsmobile and others to help them in verifying the authenticity of the information they receive on WhatsApp," adds the spokesperson.

In July 2018, WhatsApp restricted forwarding messages to five chats at a time and removed the quick-forward button for media messages in India. This was done after a number of mob lynchings were alleged to have been triggered by rumours circulated on WhatsApp. WhatsApp labels 'forwarded' messages and the restriction on them is now global.

WhatsApp recently allowed users to send suspicious messages — in the form of text, images or videos — to a "tipline" phone number. In response, the service classifies information in the message as "true, false, misleading, disputed or out of scope and (might) include any other related information that is available," WhatsApp has said in a statement.

Facebook vs fake news

Facebook doesn't remove fake or false news, but reduces its distribution in the News Feed. The company's rationale for the policy is that it doesn't want to stifle public discourse.

Facebook takes action against fake news through a combination of machine-learning technology and human review, according to a company spokesperson.

First, it uses various signals, including feedback from people on Facebook to predict if articles, photos, and videos are false. "Anyone can give us feedback that a story they are seeing in their News Feed might be false news. Feedback from our community is one of the signals that power our machine learning model and helps us take action against stories that might be false," says the Facebook spokesperson.

Facebook then sends this content to third-party, independent fact-checkers to review accuracy. Fact-checkers can also identify content to review on their own. "If they rate a piece of content as false, it will be shown lower in the News Feed — reducing future views by an average of 80 per cent. Pages and domains that repeatedly share false news will also see their distribution reduced and their ability to monetise and advertise removed."

Facebook also uses technology to match fact-checker ratings to duplicate content, so that it can catch greater quantities of false news. "For example, after a fact-checker has debunked a native image, Facebook can find additional examples of these images across the platform, and demote them," the spokesperson adds.

Facebook has the most number of fact-checking partners in India. "We currently work with seven different organisations that fact-check in eight languages — English, Hindi, Bengali, Marathi, Telugu, Tamil, Malayalam and Gujarati — and are exploring even more. In a country largely driven by local and community news, we knew it was critical to have diverse coverage both regionally and from a language perspective," says the spokesperson.

An India-specific Facebook measure is to flag a piece of content debunked by a third-party with a screen. The screen tells users what the third party has said about the content. However, users can still view the debunked content if they wish.

Twitter and truth

Twitter did not have a policy for validating content, senior company officials told Business Standard.

Twitter believed a single actor, including itself, shouldn’t judge what was true or false and taking down content because it was incorrect would undermine the open democratic debate, the officials said. It is also impossible to check every tweet.

Twitter's open and real-time nature is an antidote to false information: inaccurate content is often debunked promptly by users, including journalists and news media organisations.

Last year, Twitter said that it would remove fake accounts to protect the integrity of elections worldwide. Accounts that deliberately mimicked or were intended to replace accounts suspended for violating rules might be identified as fake accounts.

Twitter has also given users more freedom to report fake and suspicious accounts.

While Twitter does not validate the accuracy of content, company officials said that often, misinformation will be shared on the service in a way that violates its existing policies such as those related to spam -- for example, those who seek to manipulate Twitter's trending topics list will be subject to action in this way.