Separating the truth from the lies in world of online news
Pheme helping journalists to assess rumours. Katie Wright reports
Both the bookies and the pollsters predicted the outcome of the European Referendum incorrectly, but do you know who got it right? Twitter.
According to analysis from Pheme, of the 291,000 tweets where a vote was expressed, leave votes outnumbered remain significantly - and that's not even Pheme's main aim.
Named after the Greek goddess of fame and rumours, the project, which began in 2014, is designed to assess online rumours, to help journalists determine what's true and what's made up.
"Professor Rob Procter (now at Warwick) and I were discussing the manual analysis his team did with the Guardian on analysing rumours circulating during the England riots in 2011 - for example, that the London Eye was on fire," says Kalina Bontcheva, Professor of Text Analytics at the University of Sheffield, who works on Pheme.
"My background is in automatic text analysis, so we discussed how rumour analysis and detection can be automated or, at least, support be offered to decision makers to help them with the process."
The Pheme dashboard uses a set of past rumours, which are manually assessed by journalists as to whether they are true, false or unconfirmed.
Natural language processing techniques are used to assess tweets for things like positive or negative sentiment, locations, organisations and hashtags, then an algorithm determines whether that tweet supports, denies or questions a particular rumour.
"Then another machine learning algorithm tries to determine overall veracity," Bontcheva explains. "This may not always be possible with high confidence, in which case, the journalists themselves can make that judgement based on the automatically collected evidence."
Due for completion in early 2017, the partnership has brought together nine teams from universities and research groups in the UK, Germany, Austria, Bulgaria, Kenya and Switzerland, and has so far examined rumours that circulated during shootings in Ferguson and Ottawa in 2014, as well as the Germanwings crash last year.
"We have also just initiated studies of some hoaxes and conspiracy theories that circulated during the referendum, but this is still in early stages," Bontcheva says.
With news now a 24-7 business and social media sites increasingly becoming an indispensable source for journalists, Pheme could prove useful.
"Determining fact from fiction online is in many ways harder than ever, with false information able to spread far and wide within seconds on platforms like Twitter - particularly in breaking news situations," says the Press Association's social media editor, Stephen Jones.