The Power of Social Media

Data Science
This has been a tumultuous year,

with somewhat shock results in both the EU Referendum and the US Presidential Election. In the run up to both, the ‘establishment’ was largely favoured – with Remain tipped for victory by large parts of the British press and Clinton widely tipped – by as much as a 95% confidence at one point – for victory.

Yet throughout these campaigns, SPARCK’s Twitter Lab tracked substantial victories for the other side. In the entire run up to the EU referendum we tracked a large pro-Leave bias, with the same story for the US Presidential Election.

Indeed on the day of Brexit itself we got 52% pro-Leave and 48% pro-Remain tweets! This was far more accurate than any of the exit polls run at the time.

So what happened? I think we are observing two major effects:

  1. Pollster shame. As candidates and options push out to the extremities, we are seeing people potentially lie about which candidate they are in favour of. When the candidates are more middle ground – and by extension reasonable – there is less shame in openly admitting favour for them, but for the choices people have had on the cards this year that hasn’t been the case.
  2. A new battleground. I initially expected the US presidential election to be an election won by Bernie Sanders on social media – given the massive movements that were being observed on social media sites like Twitter. Instead, Donald Trump adapted his strategy well and managed to use social media to provide a platform for his followers who were often ignored and left out by mainstream media.
Word cloud for Hillary Clinton on election day
Word cloud for Donald Trump on election day

We chose to look at Twitter because it is a great source of public posts. Facebook suffers from having friend circles which strongly limit the visibility of posts you can see, whereas default account settings on Twitter are to make tweets globally available, and people – careful (particularly this year) about revealing what they really voted for to pollsters – are far happier to announce to the world their thoughts and actions.

The Commercial Power of Twitter Data is that organisations don’t need to rely on customer surveys anymore, but instead, can monitor social media reaction. If a company is trialling a new initiative, instead of just using standard KPIs to measure response, we can directly collect customer opinion and use data science to aggregate these opinions and pull out common themes. For example, in just 5 hours, we have collected over 50,000 tweets regarding a major retailer’s Christmas advert to determine the public’s honest reaction. Even some algorithmic based hedge funds are now looking at data sources such as twitter to gauge reactions to news that may impact markets.

Watch this space – in January we will be releasing our Christmas advert coverage to see who won the ‘Best Advert’ wars!

Henry Brown

Data Scientist

Henry is passionate about data and the power of prediction. He has been a Post-doctoral Research Associate at the University of Liverpool and completed a PhD in Particle Physics based in CERN, Geneva.