Fighting fake news: Decoding ‘fact-free’ world of WhatsApp

If the WhatsApp universe is to be believed, Congress president Rahul Gandhi made some startling political announcements in the run-up to last year’s state assembly elections. “BREAKING NEWS” the screengrabs of ABP news, a Hindi-language television news channel screamed, attributing the following politically suicidal statements to Rahul Gandhi:

“My ancestors were Muslims. I am a Muslim.”

“Congress is a party for Muslims and will remain so.”

“Kashmir should be given to Pakistan.”

The fact is, ABP news did not carry any such news item and Rahul Gandhi never made any of these statements.

In the fact-free world of WhatsApp, though, it didn’t matter: these fake images went viral.

A similar-looking fake image of ABP news was then circulated in pro-Congress groups to target Prime Minister Narendra Modi and the Bharatiya Janta Party (BJP): “Modi busted: 70 thousand crores rupees uncovered in Modi’s Swiss Bank account. BJP stirred”; “BJP caught in one lakh crore rupees scam, CBI reveals”.

Again, it was fake news.

Fake news that goes viral

Variants of these fake ABP news images were the second most shared misleading images in over two thousand politics-focussed and public WhatsApp groups that we monitored in the run-up to the 2018 assembly elections—offering a peek into the themes that dominated the “fake news” ecosystem.

This example perfectly represents the three key things we identified about visual misinformation in our dataset. First, newspaper clippings and television news screen grabs — real or fake — were extensively shared. Second, anti-Congress misleading content aims to create confusion about Rahul Gandhi’s religiosity (to show him as non-Hindu) and portray Congress as an anti-Hindu party. Third, anti-BJP misinformation is targeted to show Prime Minister Narendra Modi and the BJP as corrupt.

 

From outright falsehoods and partially-true misleading narratives to bigotry and hate, everything is circulated in the closed encrypted world of India’s most popular messaging application, with over 230 million active users. According to a 2017 Lokniti-CSDS Mood of the Nation (MOTN) survey, around one-sixth of WhatsApp users in India said they were members of a group started by a political leader or party.

Not all political discussion on WhatsApp is about consuming fake news. Our dataset — over a million messages collected from politically-motivated WhatsApp groups between 1st August and 4th December 2018 — has all kinds of information: long crafted text messages, infographics, political memes, and news videos. The focus of this analysis is restricted to the study of images—36% of all messages; 9% were videos.

Data collected from “public” WhatsApp groups

WhatsApp is a black box for content: It is nearly impossible to comprehensively track misinformation in the WhatsApp sphere as the content is end-to-end encrypted and no one can — and should not — access private conversations at scale. Given this restriction, we decided to monitor “public” WhatsApp groups that are open to the public, meaning anyone can join these through links publicised on the internet.

They constitute a small sample of the hundreds of thousands of groups that parties and their supporters have created to disseminate political information. We do not claim that public groups represent the discussion in private groups. But in absence of any other data, our research offers a window to understanding the themes that political actors or their supporters want to promote and distribute.

Methodology

We manually classified each WhatsApp group in our dataset with its party affiliation: of the total, we identified 693 pro-BJP groups; 156 pro-Congress; and the rest voice support for various regional parties and religious groups. To be sure, it is not known how many of these groups are managed by the office-bearers of political parties, though there is some evidence of centralisation.

For instance, around 400 groups in our database were created by just 10 phone numbers. Overall, 1,49,305 people were members of the groups we monitored. 79,781 members (53% of all members) sent at least one message; 31,459 (21%) shared at least one image.

We analysed all the images in our dataset by programmatically grouping similar-looking ones into clusters using an image hashing algorithm. Then, we manually reviewed the top 2,000 clusters (each had at least five images), together comprising around 67,000 images, and identified the misleading content in this data set. We used the shortlisted images to qualitatively study the most shared misleading images and derived the themes that dominated the fake news ecosystem in the run-up to the recent assembly elections. Here are our three key findings:

Screen grabs, news clippings to fill trust deficit

Seven of the ten most shared misleading images in the pro-BJP WhatsApp groups were media clippings. The most shared image was a screengrab of a primetime segment of Times Now, an English TV news channel, claiming that the Congress party manifesto in Telangana was Muslim-centric. Seven “Muslim only” schemes were included in the manifesto, the image claimed, including a scholarship for Muslim students and free electricity to Mosques. Except that the information was misleading. Alt News, a left-leaning fact-checking news website, later debunked how the news channel had misreported the story, by selectively picking parts of the manifesto to create a false narrative.

This message repeatedly appeared in various forms — eight of the top ten misleading images in the BJP groups were only about the manifesto — including screen grabs from CNBC-Awaaz, another news channel, and standalone graphics.

The example illustrates a key point: “fake news” as commonly understood has various shades. Unlike the morphed ABP news screenshots (second most shared) that propagated outright lies, the Telangana manifesto story is based on partially-true information that was later found to be misleading. The intent in the latter case is not clear and often difficult to establish.

Why are there so many media clippings? One possible explanation for this phenomenon is that WhatsApp -ers leverage mainstream media artefacts to compensate for the declining credibility of WhatsApp content. A limited small-scale survey conducted last year across 14 Indian states by Digital Empowerment Foundation revealed that rural Indians don’t trust messages on WhatsApp blindly, indicating a lack of trust in content received on the messaging service.

We have seen that before: Prior to 2018 Karnataka elections, fake opinion polls attributed to the BBC were circulated in both BJP and Congress groups, each publicising dubious survey numbers to showcase victory for their own side.

Anti-Congress fake news: communal propaganda

Communal polarisation was the major theme of anti-Congress misinformation in our dataset: it aims to establish that the Congress party only cares about Muslims. The Telangana manifesto was one such example. Then there is bigotry, with messages that ask people to connect the dots and rethink whether Rahul Gandhi is actually a Hindu by questioning the religiosity of his ancestors.

The narrative begins with Jawaharlal Nehru, India’s first prime minister and Gandhi’s great-grandfather. “By education, I’m an Englishman, by views an internationalist, by culture a Muslim and a Hindu only by accident of birth,” reads a quote attributed to Nehru in a newspaper clipping. Nehru never said this.

The next target in line is Feroz Gandhi, husband of former Prime Minister Indira Gandhi and Rahul Gandhi’s grandfather. The messages repeat the long-standing false rumour that Feroze Gandhi was a Muslim, in contrast with the well-documented fact that he belonged to a Parsi family.

Then, questions are raised about the marriage of Rajiv and Sonia Gandhi, his parents: “Mother and father got their marriage registered in Delhi church as a Christian couple. Son claims himself as a janeu dhari Hindu. Can there be a bigger joke?” reads one message. A fact-check by SMHoaxSlayer revealed that this claim is simply wrong.

Another says: “How can a son of a Parsi man and Christian woman be a Hindu? Have you ever thought about this?”

These rumours have been a part of the political discourse for long and are constantly fed to Indian citizens. Put this into the political context and the signal is clear: the strategy to portray Congress as an anti-Hindu party, a theme BJP has leveraged to mobilise the Hindu electorate and consolidate the vote in their favour. In fact, various images shared in the groups we monitored make an explicit call for Hindus to unite to defeat the Congress.

Anti-BJP fake news: Modi and BJP are corrupt

Anti-BJP fake news attempts to show the Narendra Modi and his party as corrupt. Clippings tagged with an organisation called “News Express” — it is not clear who runs this organisation, and if there is any real organisation with that name — ran a headline saying: “Rafale scam: Former President of France Francois Hollande says PM Narendra Modi is a thief.” Hollande never said this.

Other fake clippings make similar claims:

“RBI accepts that demonetisation was a wrong decision. It is a nine lakh crore rupees scam,” reads an image post, attributing the statement to Urjit Patel, former governor of the Reserve Bank of India.

“Modi government is the most corrupt government in Asia, says Forbes magazine”

“India goes bankrupt, according to a shocking disclosure by the World Bank. India owes 1,31,000 million dollars”

None of this is true.

Steps taken by WhatsApp

On its part, WhatsApp has taken several measures in the last year: it took out full-page ads in Indian newspapers to fight misinformation, has started labelling forwarded content and limited the number of forwards to five messages only. In February, the company said it has communicated to Indian political parties that WhatsApp “is not a broadcast platform”, “not a place to send messages at scale” and accounts that “engage in automated bot behaviour” will be banned. Globally, it has deleted two million accounts per month for the past three months on that account.

The steps are expected to add friction to use of the messaging platform by political actors, but it is unlikely to contain misinformation itself: the stream of messages and the activity in India’s WhatsApp groups will just explode in the coming months.

(Samarth Bansall is a New Delhi-based data journalist. Kiran Garimella is a Post doctoral researcher at Ecole Polytechnique Federale de Lausanne (EPFL), Switzerland. Dean Eckles, Professor, Massachusetts Institute of Technology, and Tarun Chitta helped with data collection)