Whats All the Buzz?

Blog Post Number 3, Big Data Case Study - Buzzfeed

"Whats all the Buzz?"

Popular news and entertainment social sharing site BuzzFeed earns its large traffic by publishing viral content. BuzzFeed’s team identifies trending stories and their unique characteristics in order to duplicate success in the future. For example, photos of food are popular, as are photos with guns, and the color red and women in bikinis tend to rank higher with traffic than others.
“We use two methodologies to power our analysis, identify characteristics with predictive relationship to virality: quantitative and descriptive,” Harlin said. “We have to understand how the spread of content differs by social network.” “Machine learning predicts social hits: we know what’s viral before it takes off. Regression analysis and machine learning are approaches for data analysis,” he said. However, BuzzFeed's innovation doesn't have anything to do with the content is posted, it's about how that content is tracked across the multiple sites, platforms, and services through which it is distributed. More than anything else, BuzzFeed is the prime example for the multi-platform approach to media, as much as 75% of the site's output never appears on its website at all.

How is data actually used through BuzzFeed?

Data drives BuzzFeed’s content strategy. BuzzFeed has mastered being able to predict what articles will go viral on the internet. BuzzFeed now covers news, politics, business, tech, entertainment, food, international coverage, and much more, reaching over 150 million unique visitors a month.

The key takeaways are that data can be used to optimize content for sharing through the life of every article.

Before publishing, data can help determine what to write about and how to present it.

Before an article is posted, data can helps to answer questions like what to write about. There are all these APIs out there on sites such as Facebook, Twitter, or Google. So BuzzFeed can identify things people care about and figure out how these topics are already being covered. Looking at historical data  BuzzFeed can determine characteristics of a good headline, a good thumbnail, good length for a post.

After publishing optimizing how to promote the content is also important

For content, BuzzFeed can tell within an hour or so of publishing what type of stuff we should put prominently on the home page, promote on Twitter, etc. That’s where data science can be really useful.

They also use it to train their models. The way something goes viral on Facebook now versus how it went viral before tells us what to change. We’ve gotten really good at predicting what the 10 biggest posts will be on a given day. The one thing that is hard to predict is the magnitutde.

One of other things we do is try to make data accessible to our writers—give them feedback on how content is doing in a consistent and regular way. That way, we see the same things, and can try to figure out why does well on certain platforms and things like that.

What is the problem, if any is plenty of news. We believe in mobile. People don’t engage on mobile news as much as on desktops.In order to bridge the barrier of time, Circa technologists use Big Data to serve bite-sized chunks of only new news that readers are following, instead of expecting story followers to re-read content they have already consumed. Each atomized chunk is called a card.

“We strive to present each article as atomic content. It’s also context. Here is how we construct mobile news on Circa. We don’t use auto summarization; we have actual journalists.” Metrics are an important element for journalists. “You have to decide how much to give to journalists. You want to give them a nice balance of analytics that matter most to you: users, pages, sources, editorial statistics, story sharing. The metrics that mater include explicit and implicit user behavior,” he said. “We tend to look at follows and shares more than anything else.” BuzzFeed employees urge publishers to re-think their content, change the way to re-formulate it, and make it more efficient and dynamic. Big data can be used to assist in streamlining news feeds, as well as creating actionable audience metrics for your team. In a nutshell, BuzzFeed has been building a giant engine for understanding content flows and sharing behavior online, thanks to tools like Pound and The Hive—a  huge, real-time database that react-text can ingest and track BuzzFeed content wherever it is shared across multiple platforms. In summary we see how data helps determine what to write and how to promote the content. However data has yet to tell us why people share some articles versus others. That is where quantitative shifts to qualitative  Through reading comments and reading articles about articles BuzzFeed can learn where they went wrong and right.  

http://reutersinstitute.politics.ox.ac.uk/sites/default/files/Big%20Data%20For%20Media_0.pdf