Earlier this month, a 16-year-old lost his finger while dancing at a rave in a disused postal sorting office in London. He continued to dance, minus one finger, for another half hour to get value from the £10 he had paid to get in, before eventually being persuaded to go seek medical attention.

Later, he cheerfully Instagramed a picture of his pinkie-less hand from his hospital bed, which was soon picked up by the Independent in a story titled ‘Teen loses finger at Croydon rave, continues dancing ‘because the bass was hard‘.

Within an hour of the story being published, it had attracted over 1,600 interactions on Facebook.

Screenshot: NewsWhip Spike

Two hours later, the Facebook interaction count was up to almost 2,800, along with thousands of tweets. One week on, the story is the fourth most-shared piece in the UK, as well as the Independent’s biggest story by some distance, with over 60,000 Facebook engagements.

The Independent’s digital staff noticed the impact early on.

Social Media Editor Felicity Morse, who was responsible for posting the story to the Independent’s social media fans and followers, says speed was a crucial factor in the story’s success.

“It was obvious it was a great story right from the get go. You can’t really fool social media, it has to be a good story!” she told us.

“I like to think our story took off because our Facebook and Twitter followers are really engaged, they like sharing what we post and also because we sold the story on social in a way that reflected the way it was written and the way people were reading it. We were astonished by it and so was everybody else.”

Sharing in the first hour of the story – far higher than usual score for the Independent – had given some indication that this would be a big story. But could the scale of the early numbers actually predict how big the story would eventually become? Or perhaps the social velocity – the rate at which it spread in its first few minutes – held the key?


Accurately identifying viral stories before they actually go viral is no easy task.

But for newsrooms and other publishers increasingly focussed on maximising the spread of their content, it can be extremely useful.

Homepage and social media editors stand to attract huge numbers of new readers by understanding which of their stories is performing strongest, or spotting what is taking off for competitor sites.

But what’s the most reliable way of measuring viral potential?

Apart from Twitter quietly initiating a project to predict viral tweets, there has been little work by social networks themselves on predicting what stories that their users are sharing are about to go viral.

A June 18 Washington Post story on the news that the US Patent Office is to cancel the Washington Redskins’ trademark was tweeted over 3,200 times within 39 minutes of publication.

It was the biggest story of the week for the Post, attracting over 96,800 Facebook interactions and 8,000 tweets by June 25. So, could you have predicted that 39 minutes in?

The Speed of Human Sharing

At NewsWhip, we track and rank stories using Social Velocity, attributing scores to stories based on how fast they are spreading on Facebook, Twitter, and LinkedIn. Our dashboard (Spike) shows which stories are picking up shares fastest in hundreds of categories and sub-categories.

For some time, users of Spike – mainly journalists and editors – have told us that it has an uncanny ability to predict big stories, and find big stories early in the viral cycle. We wondered if we could quantify this effect.

So in February 2014, we opened up the API to a team of researchers at the Irish Centre for High-End Computing (ICHEC). The ICHEC has access to incredible computing power, and the team has advanced degrees in computer science and statistical modelling.

The ICHEC team analysed around 140,000 stories that passed through Spike’s 1 Hour View – stories published in the last one hour only, that are starting to show signs of virality. The team looked at the Social Velocity score NewsWhip attributed to these stories – the shares per hour we calculate (it runs from zero to a maximum of about 10,000).

First, the team analysed a set of the stories in the top percentile (the monster successes), the successes (the decent sized stories), the normals (the typical social stories) and the duds (the stories that did OK, or did not take off at all), based off NewsWhip data over 24 hours. The monsters are the stories that make it into our 24 hour view in Spike – these are the most viral stories in the world from that time period. The successes are still kicking after 12 hours, the typical stories survive 3 hours. The duds disappear.

The ICHEC team then calculated threshold Social Velocities that correlated with stories making it into the big leagues, using mean scores of velocities attained by historical viral stories to build the model. The team then returned to NewsWhip’s live stream of new content and looked at the social velocity scores attained by stories in their first hour. This allowed the ICHEC “to determine how many of the stories that have been picked out as ‘viral stories’ in Spike 1 Hour did make it to 24 hour,” their Data Architect Dr. Bruno Voisin explained. Spike’s 24 hour view shows the biggest stories of the last day across Facebook and Twitter.

Early Velocity Predicts Later Results

The research team found “a high degree of predictive accuracy in Spike 1 Hour, identifying around 79% of the biggest, most viral stories – which were evident in Spike’s 24 hour model.”

Stories that would achieve mid-size viral success were also accurately picked out, with lower scores. “The results do suggest that the social sharing around a story in its first hour after publication is highly predictive of how big it will eventually become at a later stage,” said Dr. Kashif Iqbal, Senior Software Developer at ICHEC.

The top level number is fascinating: 79% of super-viral “monster” stories exhibit an outstandingly high social velocity in their first hour. Similarly, stories at other lower velocity thresholds fall into the merely “successful” or “normal” categories, as appropriate. So merely fast moving (not speeding) stories will tend to get mid-range sharing, and slower stories tend not to spread further. Early trends tend to follow through.

For example, this story with a current velocity in the low hundreds is destined for a mid-size social success, with a score of 298. (See the velocity on the right side of the story, marked “Current”.)

On the other hand, Will Farrell’s inspiring pep talk (“I’ll bite every German Player if I have to”) looks set to be a viral hit for the ages with a current speed in the thousands. This one is likely to grow and grow.

This is a useful finding for anyone in the content creation game. For editors and writers, using hard social metrics to watch the web for viral hits might beat guesswork on picking winning stories. (This is effectively what our customers use Spike for today.) An additional benefit for publishers is that early sharing signals can warn of a slew of likely traffic.

Finally – this study showed that only 79% of stories exhibit these “early growth” characteristics. Explaining the other 21% of stories is interesting – these are stories that go viral, but don’t get a high score in their first hour after publication.

It could be that some stories don’t immediately get noticed, or don’t get traction until they are amplified by a bigger media company, discovery by an influencer, or highlighted by a curator of viral content like Upworthy. Perhaps they are published while their target “sharing market” is asleep and only get shared when it wakes up. And of course, some content – like quizzes or listicles – might just gain slow and steady growth on the social web through less excited, slow and steady referrals. But it seems from our data on big social hits, that those “slow and steady” stories are in the minority. Big stories get big quick.

If you’re interested in finding viral stories early in their cycle, sign up for Spike today