US election provides new benchmarks for social media activity

By Nic Newman Senior Researcher City University, London

Every day more and more content is created and distributed through social media, but the key problem for consumers and professional journalists alike is how to find the good bits.

The Social Sensor project is looking to find new ways of solving this filter problem. What algorithms can identify the best tweets, pictures or videos? How can we separate truth from lies in the heat of a fast moving news story?

Many of these issues were laid bare on election night (November 6th 2012) when in addition to traditional television and web coverage the drama of the night was also played out on Facebook and Twitter. We set out to monitor the range of conversations and identify the kind of information that could have been useful to professional journalists

The key trends

The election was won by the Democrat incumbent Barack Obama who defeated Republican Mitt Romney with 332 Electoral College votes - compared with 206. The Democrats retained control of the Senate, while the Republicans held the House of Representatives. Results were declared state by state from 7pm Eastern Standard time with key battleground states of Ohio, Florida and Virginia taking longer to emerge than expected because of long queues and the closeness of the race.

Overall, 30 million tweets were sent on election day making this the most tweeted political event in U.S. history. Barack Obama's victory announcement was done with a tweet – which became the most retweeted in history and the most liked post ever on Facebook.

Twitter said the total tweets sent on election night averaged about 9,965 per second for a full hour following Obama's announced win, with a one-second peak of 15,107 tweets per second and a one-minute peak of 874,560 tweets per minute.

An election is an odd story, typical in some ways but not in others. Most of the comments, pictures and videos that were posted and shared tended to pick up from what mainstream media or politicians were saying. Social media didn't break stories on the night but rather acted as a space for commenting on and amplifying what was unfolding on television and the web.

This is partly because big news companies had privileged access to results data supplied by AP (Associated Press) and their own polling data from which these networks made the projections of a victory or delivered analysis

This is very different from an 'uncontrolled distributed news event' like Hurricane Sandy where much of the news itself came from social media – or a breaking news story like the Connecticut school shootings where journalists were to be found scrambling through social network pages for eyewitness accounts.

Overall, through content analysis and observation, we picked up three core use cases for how journalists were using or could have used social media on election night.

  - To pick up 'unexpected' trends from the big social data that the media hadn't been able to see (e.g. voting lines and stories emerging)
  - To quickly spot news lines from politicians that were released first via Twitter (e.g. Barack Obama victory tweet, Jeb Bush saying Florida had been lost etc.)
  - To provide colour, quotes, pictures and context for live blog pages and other user generated content (UGC) that networks and newspapers were incorporating into their coverage

In all three cases, the challenge is to provide an interface that could filter the mass of real time data and surface the most relevant content in a way that suited the interest of journalists.

Social sensor trend analysis

As well as monitoring a range of tools and documenting a timeline of events, we also monitored social media in real time using a tool developed by City University to pick out topic clusters. This tool took content from a Twitter seed list of 5000 key influencers on US politics. The basic idea is that if we could listen into the conversations of these committed 'news hounds', we wouldn't miss anything important.

Overall we crawled 37 GB of tweets in around 50 hours. Our initial analysis used just 12 hours of tweets. Our 5000 "US politics" accounts delivered around 3,881,218 tweets. The tool processes and clusters tweets in near-real time with updates every two minutes.

The picture below gives an idea of what this tool looked like just before victory was called at about 3.40 GMT.

This tool identified the key issues quickly, but as the picture illustrates many of the clusters just highlighted news that had already been reported on TV (eg Romney wins Arizona). A tweet by Al Gore (also illustrated) predicting that President Obama would win Florida was more significant and of great interest to journalists – because Florida is where Gore lost the election to George Bush in 2004.

Political tweets were widely monitored, quoted and used by the mainstream media. Notable examples included Republican Jeb Bush's acceptance that Obama had won Florida and Anne Romney's exclusive pictures of her husband watching the results come in (see below).

During the campaign, New York Times data expert Nate Silver had become something of a celebrity with his deep insights on the state of public opinion. His tweet 'On The Wall, The Writing' was extensively retweeted over the next few hours (see below) and reported on TV and radio – taken as a signal that the election was over and illustrating the power of particular journalists to influence events.

Most popular trends on social media

Here is a list of some of the other key trends that emerged through the night on Twitter

  - The battle in Ohio, Florida and Virginia as the three key swing states
  - The victories as they came in the states that declared very hour – including the demographic information that emerged from each
  - The high turnout generally (favouring Obama) and specifically with Afro Caribbean vote
  - The Hispanic vote and its importance
  - The women vote and how it was heavily for Obama (and the male vote for Romney)
  - The queues in line to vote – up to 2 hours - and the appeals on Twitter by each camp to stay on and not go home (see below)
  - Various state races especially Elizabeth Warren's victory in the Senate
  - Republicans fail to take senate – Democrats do better than expected

One of the most interesting developments was that the political parties were monitoring developments in key states and then using Twitter and other social networks to try to get their supporters to the polls. The Obama campaign was quick to react to the long queues (lines) in Ohio and to reassure them that it was worth staying in line to vote using social media. This message was extensively retweeted by Democrat supporters as the images below illustrate.

Our Social Sensor trending tool also picked up this viral message

This further chart shows how the Obama team's viral messaging kept swirling around Twitter at a high level for a number of hours.

One interesting development earlier in the day was the number of pictures of ballot papers that were posted on social networking sites like Instagram and Twitter. This sparked a conversation about the legality of taking cameras into the polling stations.

According to the Pew Research Center, 22% of registered voters indicated on Facebook or Twitter who they voted for. Google's YouTube also encouraged people to document the process and CNN allowed you to express your preference via a hashtag campaign, which then lit up the Empire State building.

Mainstream media coverage

Crowdsourcing social media reactions, comments and pictures were a key part of mainstream media coverage. Many of the big newspapers such as the New York Times and the Wall Street Journal ran their own live TV style coverage on election night for the first time – as did the Huffington Post which mixed the video footage with Facebook and Twitter comments from users.

Live blogs (sometimes known as live text commentary) were also the default way in which websites covered the story, together with live data maps showing the results as they came through. The picture below shows the BBC live stream, which extensively used social media comments to add context to the real news.

The New York Times and many others pushed the boundaries around live blogs adding live results modules alongside the traditional news + social media mix.

Conclusions

Social media has now become a core part of the toolkit of mainstream media – which routinely monitors and uses this content as part of its output.

In this particular story, the most important element was the picking up of politicians' comments by journalists. For the first time in an election we saw politicians deciding to use this channel first as a way of getting their messages out. Twitter was the most effective way of doing this because of its speed and the open nature of the network. For the first time, we also saw TV pundits following trusted commentators via their Twitter feeds and relaying that information live on air (e.g. Jeb Bush, Al Gore tweets).

In addition to changing journalistic practice, social media has brought up new questions about how open voters should be about the electoral process. Voters are now taking cameras into the voting booth and telling other people how they voted in a way that simply didn't happen four years ago. We can expect these trends to continue and for journalists to try to use this trending to inform coverage of many types of story.

Finally, we have seen politicians becoming more sophisticated in using social media to target key voters in battleground states while polling is going on.

For all these reasons the importance of monitoring and identifying real time trends in Twitter and other social media is set to grow – as will the need for every more reliable tools.

Nic Newman is a Senior Research Fellow at City University and an expert on Digital and Social Media. He is former World Editor of the BBC News Website and a former Controller of Future Media for BBC Journalism.