x Abu Dhabi, UAEMonday 19 February 2018

The Twitter software that knows how you feel

Imagine a software program that analyses the way the entire Twittersphere is feeling about any given topic – and in so doing gives those in charge the power not only to control those feelings but also to manipulate the events to which they might lead.

In living memory, emotions used to be private - we kept them to ourselves or our closest friends and acquaintances. The internet has changed all that, of course, offering each of us a voice and platform from which to broadcast our surprise, shame, sadness, etc on every topic from the successes and failures of our favourite sports teams to the performance of the government.

But now imagine a software program that analyses the way the entire Twittersphere is feeling about any given topic – and in so doing gives those in charge the power not only to control those feelings but also to manipulate the events to which they might lead.

Welcome to Emotive – Extracting the Meaning Of Terse Information in a geo-Visualisation of Emotion – a program developed over the past year by the Centre for Information Management at Loughborough University.

At its heart is the exhaustive business of teaching the program the many words in the English language to convey a given emotion - there are, for instance, about 50 ways of expressing hate.

If you think Emotive sounds like an Orwellian development courtesy of the Thought Police from the dystopian novel 1984, you aren’t alone. Even Tom Jackson, the professor of information and knowledge management leading the team at Loughborough, concedes that “a mechanism to intervene and potentially change the way the world is actually thinking is quite scary”.

Emotive, he agrees, could be used for good or evil.

“For the good, well, retrospectively we mapped the 2011 London riots and you could have seen it was escalating out of control and that some intervention was needed,” he says. “That might not have been law-enforcement intervention, it could have been talking to leaders of particular gangs and trying to quell the disturbance.”

Likewise, the team used Emotive to map the development of public feeling over the murder in May of off-duty British soldier Lee Rigby, hit with a car and hacked to death on a London street by two Islamist extremists.

“You could see anger and disgust was growing at a phenomenal rate,” says Prof Jackson. “Then the family came out and said some nice, kind things; this introduced a bit of happiness and started to quell the anger, disgust and sadness, and the map started to revert back to much more normal levels.”

And the system can learn to predict how events will unfold.

“So if you start mapping these events and there’s another brutal murder – and I’m sure there will be, because that’s life – instead of it escalating and turning to riots, you could potentially get to the key people to get out some messages that could potentially quell the situation.”

The question, says Professor Jackson, is “how far do you go, in terms of Big Brother? If you determine that you can change the course of events – that you could potentially change the course of an election, for example – would that be right or wrong? You are getting real-time feedback on how your campaign is going and you could look at doing smear campaigns and see how that affects the opposition.”

Monitoring Twitter sentiment is already big business.

This year, Sky News Arabia, the new kid on the 24-hour TV news block in the Middle East, turned to social analytics to develop what it called its “sense and respond” news model, monitoring reactions not only to its own tweets and blogs but also those of its competitors and reacting accordingly.

To achieve this, it turned to Brandwatch, a new company offering software designed to fillet every single scrap of useful intelligence from the digital ether. The upshot, said Fares Ghneim, communications manager for the company, a collaboration between Abu Dhabi Media Investment Corporation and the UK’s BSkyB, was that “we feel the love or the hate very quickly – instantly, in fact”.

Setting aside the problematic question of potentially shaping the news to satisfy what could well be the prejudices of a small but digitally active proportion of viewers, the model is also a tempting one for politicians. As the promises made by the opposition Labour party in the UK at its annual conference last week demonstrated, modern democratic politics is more about achieving power by giving people what they want, rather than what is necessarily good for them.

Twitter itself is already highly tuned to the possibilities, as the activities of its Government and Politics Team demonstrate. The “explosive growth in conversation” it celebrated last year had “fuelled Twitter as a platform for civic debate and created a massive data set for analysis”.

Accordingly, in August last year, in the run-up to the US presidential election, it introduced its Twitter Political Index, “a daily measurement of Twitter users’ feelings towards the candidates as expressed in nearly two million Tweets each week”.

It quickly became apparent that no politician’s team could afford to ignore Twitter’s metrics. Of course, measuring the sentiments of only the Twitterati is to focus on the feelings of a self-selected group and ignore the rest of the population, who fail to engage with social media for a range of reasons – including disinterest and political indifference.

Nevertheless, the sheer scale of the Twitter data flow is breathtaking: according to the organisation in 2012 more Tweets were sent every two days than in the entire history of Twitter prior to the 2008 election. Put another way, the entire output of Tweets related to the 2008 election was matched in any six minutes of Tweets before the 2012 election.

Today, the Twitter Political Index – #twindex – is a constant political presence, computing a live, rolling approval rating for the president by analysing about two million Tweets a week. And just how accurate it could be was made clear when Twitter published a chart comparing its analysis with that of Gallup, the traditional polling company, and demonstrated similar peaks and troughs of sentiment, but which much finer attention to detail.

Indeed, during last year’s State of the Union Address, Twitter’s government and politics team spewed out sufficient analysis, backed by huge sample numbers, to drive even the toughest polling veteran to hang up their pen and clipboard.

Between 9.05 and 10.40pm – from the moment Barack Obama stepped up to the mic to the end of the Republican party’s response – Twitter analysed 766,681 live Tweets. Compare that with a traditional polling sample of 1,200 voters.

If the president was uncertain as to his future policy direction, he had only to consult the top-five core hashtagged themes that emerged as major concerns for the tweeting viewers – education, energy, innovation (including a politically deft, top-scoring reference to Steve Jobs), fairness and manufacturing, while analysis of the top tweets by members of Congress also revealed the chief concerns of Democrats and Republicans alike.

For some, such close attention to the minute-by-minute whims and tweets of a digital elite suggests a slide towards a dystopian model of populist politics, in which politicians pander to the section of the mob that “shouts” the loudest in a way that has never been so possible before.

Such a development, if taken to the logical extreme, could signal the final, cynical victory of political expediency over belief. Consider a presidential speech: the president, monitoring feedback on a discreetly placed tablet, could even be guided by his advisers to boost his ratings by changing tack on the hoof.

The commercial applications, though no less cynical, are at least less sinister.

“Take The X Factor as an example,” says Professor Jackson. “If you are doing a live show and you see people are getting fed up with the acts, you could shake it up and put on a different act. And when it comes to advertising, you know what sort of state the nation’s in at the moment, so you could actually be quite precise with the type of advertisement you want to put out to make the most of those emotions at that time.”

It seems unlikely, however, that the UK Ministry of Defence is interested in manipulating the outcome of the next series of The X Factor. It is now taking the research to the next level, recruiting five universities, including Loughborough, to develop strands of a project it calls Redites – Real-Time Detection, Tracking, Monitoring and Interpretation of Events in Social Media.

Prof Jackson sees the Middle East, with its fast-moving landscape of protest and revolt woven inextricably into the ebb and flow of conversations on social media, as a natural opportunity for Loughborough’s Emotive, and says he would welcome a collaboration with a university, or a government, in the region.

“I think it would be quite powerful,” he says. “It would be great if we could use it in the Middle East and start to see the feelings of the people.”

It would, he says, be necessary to recalibrate the program to take account of regional semantics, and that’s where local linguistic expertise would come in.

“You would have to do quite a bit of surveying, because emotions would be very different to the UK, so we might be reacting to things that are just normal passion in that part of the world. You would need to find the baseline of what is OK, and reasonable.”

Whether it is reasonable to take a nation’s emotional temperature in this way, and to act to cool it down when it threatens to rise too high, is probably a moot point – after all, Twitter is a voluntary, public platform, used by increasing numbers of people precisely because it allows their voice to be heard.

But Twitter, says Prof Jackson, is not the only potential application susceptible to Emotive probing.

“If you think of all the data the NSA now has access to in America, you could use it on that, and that worries me.”