Twitter – UVM researchers create trend-prediction software
BURLINGTON, Vt. (WCAX) – An estimated 500 million tweets are sent out on Twitter in a single day. Now, new software developed by University of Vermont researchers helps people catch trends sooner.
“Why we see the patterns in the world that we see,” said Chris Danforth, a UVM professor of math and statistics, who helped launch the StoryWrangler as part of a doctoral study. “What people are thinking about, how information and misinformation, and disinformation are spreading.”
The prototype software uses fast computers at UVM that read every single message, tally up how often words or phrases are used, and show the patterns. What’s online can translate to social movements ranging from Black Lives Matter to the Capitol riot. “Try to understand how those movements coalesce, where people are getting their information — those are really important applications that we see for this,” said Danforth.
Every time you open Twitter, whether it be on your laptop or your phone, you get a personalized version of the site based on your data, but with StoryWrangler, you get a whole new perspective, something you may not have looked at before. “It’s real-time, it’s across the globe, it’s different languages, different cultures, different socioeconomic backgrounds, so it’s sort of like the filters are being removed,” said David Bradbury, president of the Vermont Center of Emerging Technologies. He says knowing exactly what people are talking about has value. “If you can understand the conversations in near real-time — or real-time in this instance — you should have better insights and insights lead to actions.”
Actions like health care challenges, resource distribution, money-making, brand management, or even just what’s trending. Bradbury says it can’t replace sample or focus groups, because Twitter has voluntary usage, so not everyone is signed up.
Danforth says they still have hurdles to jump. “We can provide valuable real-time information while also preserving the privacy of individuals that may not want to be a part of a dashboard like this,” he said. That’s why he says they’re making re-identification nearly impossible. And since this is a prototype, there’s more room for growth around their central idea. “Better information about how people are feeling and doing.”
Danforth says the prototype is only testing on Twitter and that while the social media platform isn’t the perfect sample, it definitely is a large one. “One of the long-term applications here is to sort of build an unsolicited public opinion poll on anything you can think of. So, you go on the website and type in a word or phrase that you want to know how people are feeling about it, what sorts of words are people using to describe it, what stories are being told about it — and it can be something as important as the president or as mundane as a breakfast cereal — and then you are able to see how people are talking about it in the wild,” said Danforth.
He says the biggest concern so far is simply data usage and privacy.
Copyright 2021 WCAX. All rights reserved.