During a recent American political convention, two networks carried real-time Twitter 'data', what they called a meter, based on the commentary on the social media service.  While election polls have not been accurate for the last few decades, some technologists believe Twitter can be.

As evidence, they say the have devised a means to predict the outcome of other election-based processes - TV talent shows - through their big data analysis of tweets. Analysis of Twitter is not new, it is built into their API, but Fabio Ciulla from Northeastern University in Boston, USA and colleagues say their method shows that the elimination of contestants in TV talent shows based on public voting, such as "American Idol", can be anticipated. 

The authors used the voting system of these shows as a basic test to assess the predictive power of Twitter signals. They relied on the overlap between Twitter users and show audience to collect detailed data on social behavior on a large scale. This approach provided a unique opportunity to apply network science to social media and study social phenomena in a completely unobtrusive way. Previously, Twitter had been used to forecast epidemics spreading, stock market behavior and election outcomes with degrees of success that were more luck than science.

In this study, the authors used Twitter activity during the time span limited to the TV show airing and the voting period following it correlated with the contestants' ranking. As a result, it helped predict the outcome of the votes. This approach offers a simplified version helping to analyze complex societal phenomena such as political elections. Unlike previous voting systems, Twitter offers a quantitative indicator that can act as proxy for what is occurring around the world in real time, thereby anticipating the outcome of future events based on opinions. If you accept that definition of proxy, anyway.

Ciulla and colleagues say that the fraction of tweets that included geolocalization information made it possible to internationally map the fan base of each contestant. They identified a strong influence by the geographical origin of the votes, suggesting a different outcome to the show, if voting had not been limited to US voters.

Published in EPJ Data Science.