Tweets analysed for clues to UK general election result

Analysis of Twitter comments could provide an accurate forecast of the UK general election result, computer scientists from Warwick university said on Tuesday.

About 160,000 tweets a day, excluding retweets, are being posted on the topic of the general election. A predictive algorithm, initially developed in Greece and refined at Warwick, seeks to define and extract meaningful features from the vast quantity of user-generated content, harvesting political tweets to make faster and more accurate predictions of the vote share.

The method, which includes negative as well as positive tweets, aggregates this information with the latest estimate of the parties' share of the vote as measured by conventional opinion polls to produce a daily prediction of voting share.

Tuesday's prediction shows the Conservatives slightly ahead of Labour with 33.48 per cent of votes compared with the opposition's 33.06 per cent of votes. The data also suggest an improvement in the Liberal Democrats' vote share since May 2 to 10.6 per cent and a slight weakening for the UK Independence party to 12.19 per cent.

The tool also provides charts showing the level and nature of discussions on Twitter since March 21 about the six main UK political parties.

The Warwick team said analysing tweets allowed them to spot changes in the public mood quickly. This, they suggested, could be particularly useful just before election day when much opinion polling will have stopped.

"We think that Twitter is very noisy," said Adam Tsakalidis, developer of the core of the prediction algorithm and one of the team based at Warwick university's Department of Computer Science. Tweets may be spam or mainly negative comments about a particular party. But "the best advantage is the vast amount of users tweeting almost in real time," he said.

This does not overcome the objection that Twitter users are not a cross-section of UK voters. But by combining the computer-based Twitter analysis with polling data, the team believe they can achieve more accurate forecasts.

The approach, developed in collaboration with the Department of Journalism at City University London and the Information Technologies Institute in Saloniki, Greece, achieved better results than other polls in the Greek election in January, Mr Tsakalidis said, and in the 2014 EU elections in Greece, Germany and the Netherlands.

"Predicting elections using social media data has been tried in the past, with varying results," he says. "We have evidence from our previous work that our approach is very effective."

© The Financial Times Limited 2015. All rights reserved.
FT and Financial Times are trademarks of the Financial Times Ltd.
Not to be redistributed, copied or modified in any way.
Euro2day.gr is solely responsible for providing this translation and the Financial Times Limited does not accept any liability for the accuracy or quality of the translation