Blog

Forecasting the US Presidential Election

Posted: 6th March 2012
By: CHRIS

Today is Super Tuesday in the United States, which sees the most delegates up for grabs during any single day of the Republican nominating process. With that in mind, we’re excited to share our new election monitoring application that uses sentiment analysis to forecast the GOP presidential nominee. Ultimately, we’ll shift the site to predict the general election winner.

Recorded Future analyzes content from more than 150,000+ public sources in real time, and processes hundreds of thousands of pages of text looking for information about the future. In essence, we’re recording what the world knows or believes will happen, structuring and scoring that information, and then making it available for analysis.

In this example application, we created a custom score called Win Sentiment to reflect the web’s feeling about who will win the Republican nomination. Win Sentiment is based on a number of factors including the language used to describe each candidate, weight of the reporting sources, and beliefs about each candidates’ future. As the tone related to each candidate’s chances of winning turns more positive or negative, the predicted winner is updated in real time.

At time of writing, Mitt Romney leads no matter how you filter the data, but it will be interesting to see if sentiment expressed during the course of Super Tuesday shifts the predicted winner. Also, there are interesting patterns to be found using the media and location filters such as the favorable shift for Ron Paul (compare below image to the one above) when you select “Niche” media versus the general coverage.

We hope you’ll enjoy using the site to watch news flow about the candidates in real time, see whose fortunes are rising and falling in the eyes of the web, and learn who the web thinks will be the overall nominee. Check back periodically, and especially after the GOP nomination process concludes when we will tune the dashboard to predicting the general election!

Related