Home News GA Tech Ph.D Simplifies Your Social Feed

GA Tech Ph.D Simplifies Your Social Feed

by Muriel Vega

Every second, on average, around 6,000 tweets are posted on Twitter. That’s around around 200 billion tweets per year. It’s a miracle we can even keep up with our feeds, especially during a major event like the Super Bowl or an election. What if you could visualize it all at once and process the information more effectively?

Mengdie Hu, a Georgia Tech Ph.D. graduate in Human-Centered Computing, began analyzing possible solutions after a summer internship at Twitter.

“I was interested in visualizing texts, especially related to social media,” says Hu. “At the time I thought, there could be more effective ways of visualizing social media text for people to analyze the data.”

Instead of endlessly scrolling through a timeline to get the gist of a major event, SentenTree (short for Sentence Tree) presents your information in short text. It allows you to engage with the conversation and interact with the most talked about words from an event in minutes. Hu’s project works best during heavy interactions times like an election, major sporting event, or latest product release (think Apple’s massive product launches).

“Given a large collection of tweets, can they be summarized in one concise representation that gives the viewer an idea of the key themes and ideas throughout the collection? Thus, the viewer can get a good sense of what the key topics are in the collection at a glance, without having to read many of them,” says John Stasko, professor of Interactive Computing at Georgia Tech and part of the research team.

In a demo, almost a quarter of a million World Cup tweets posted over 15 minutes are filtered through the analytical program. The end result is similar to a word cloud, where words that appear in tweets more frequently show up larger — in this case, Brazil, own goal, soccer, etc. “Unlike a word cloud, particular keywords can appear multiple times in the display because that word may appear in different locations in different tweet/sentence fragments,” says Stasko. If the user would like to dive deeper, you can click on a chosen word and reach a more detailed part of the conversation.

“SentenTree itself, it’s what we call a technique rather than a tool,” says Hu. “It’s more like an algorithm. We can actually plug this into other analytics tools, and we’ve actually got a few requests after we published a paper of people interested in adding SentenTree to a kind of suite of tools to analyze social media text.”

Hu and her research team hope to open up the algorithm source code soon for professionals to pair it with other analytical tools and for recreational users to add their own tweets or datasets.

You may also like