News

  • RSS
  • Print

Big Data, Social Media, and the Elections

Obama's Facebook Fans Love Michelle, Romney's Love Winning – that’s how Rebecca J. Rosen summed up American University School of Communication professor Deen Freelon’s research on responsiveness to Facebook posts made by the two 2012 presidential campaigns.

The campaigns’ use of social media in the 2012 election has been discussed by pundits and analyzed by the campaigns themselves with an eye to strategic efficacy before and after election day. But Freelon’s research has a different focus – what individual posts truly resonated with each candidates’ followers?

Freelon presented preliminary results from his work as part a brownbag series held by the Center for Research on Collaboratories and Technology Enhanced Learning . These findings, which inspired the Atlantic article referenced above, have also been published on Freelon’s blog What resonated with Obama and Romney’s Facebook Followers?

What the data showed, in short, is that Romney supporters tended to be most engaged when rallying around a specific call to action, such as “Help us get to 7 million likes!” Meanwhile, Obama supporters responded most strongly to personal posts, particularly those referencing his family. He suggested that there are a number of possible explanations for this. One is that the Incumbent v. Challenger – Obama is a known entity his followers are invested in him as a person, not just a campaign. Perhaps it was due to the fact that Romney seemed to be struggling in the polls, which could lead his followers to demonstrate his popularity in online shows of strength.

Freelon’s area of research interest is mediated political expression and participation. For his current research, Freelon pulled data from the Facebook API for the Romney & Obama Facebook pages & twitter feeds from the time Romney was named the official GOP candidate through the election, using a scraper he had built himself. Freelon explained that a site’s API is like a data spigot, spewing tons of relevant and non-relevant information, and the scraping program acts like a cup to collect the specific data a researcher needs.

Additional questions Freelon hopes to tackle on the election 2012 social media data he has collected include:

  • Which online content resonated most with attentive citizens
  • What was the function of Twitter during debates
  • What message quality (bashing the opponent, praising self, call to action, etc.) best predicts interactivity

To get at these specifics will require a more qualitative analysis of the data. This includes using a variety of content indicators to learn more about the over 800,000 posts. This volume would be prohibitive if a human was required to analyze each post, so Freelon is employing supervised machine learning, in which his graduate assistants teach a computer program how to do enhanced text analysis. Some of the challenges Freelon identified with the process included figuring out whether there is a way to teach the program to find and identify sarcasm or humor in posts.