Software Prunes Twitter Tree in Real Time

Editors, dry your tears. Scientists have proposed a way to marry the credibility-striving values of traditional journalism with the real-time, first-responder vigor of the microblogger tool Twitter.

New software targeted for use by media organizations will “prune” bursting Twitter trees as they happen, deciding what accounts to promote or eliminate in a fraction of the time required by human curators.

“A key challenge in the real-time Web, which we consider Twitter a large part, is the huge amount of information being generated all the time,” said researcher Pádraig Cunningham at University College Dublin in Ireland. “The challenge we see is to identify users who are authoritative commentators.”

Breaking new stories such as grassroots unrest, natural disasters (Japanese tsunami, Gulf Oil spill) and public tragedies (the death of Michael Jackson, the shooting in Norway) increasingly develop in high-volume Twitter messages, or “tweets,” that accumulate around a topic.

But even with laborious filtering by news workers, monitoring news on Twitter “risks incomplete coverage” and  the drowning of relevant voices by those “with no proximity (in space, time, or expertise) to the news story,” researchers stated in an Oct. 6 study entitled “Supporting the Curation of Twitter User Lists” that was published on the research website arXiv.org.

“In the Bahrain context, an authoritative commentator is someone who’s early with news, who’s on the ground, and whose opinion is more insightful and more valuable than other commentators,” Cunningham said of the tool’s use on networks covering the current evolving Arab revolt.

A test of the tool’s effectiveness for a Twitter network covering the 2012 presidential caucus in Iowa found its recommended list of commentators largely mirrored that culled by humans.

“In the Iowa scenario, the evidence that a commentator is authoritative is that they have a lot of followers, someone with thousands of followers. We consider that to be evidence that their opinions are authoritative,” Cunningham said.

Previous automated models for sifting networks tended to locate only “central actors” and Web pages, Cunningham said. A search engine like Google relies mainly on “link juice” — how often a story is linked to from other quality sites — to identify results. But social media’s inherently richer zones of relationship among users and content, as well as oft-important voices from the edge, have frustrated attempts to base rankings on one perspective.

“We’re able to show that . . . by looking at multiple types of relationships [retweets, lists, and followers relations], you get a more robust result,” Cunningham said. In addition, “Twitter users may mention (e.g. @JoeBloggs) users of a different view without 'following' them. So 'mention' links may be [more] effective than 'follows' links for getting out of information silos.”

Identifying counter-perspectives is the software’s current research challenge.  “There was a famous study done in the early days of the Web. It showed that Republican bloggers and Democrat bloggers don’t communicate with each other at all. They don’t comment on each other’s blogs, they don’t link with each other,” Cunningham said. “And we have evidence that the same situation is in happening in Bahrain, but the polarization is even more extreme.”

Developers expect that the software will always require at least some human intervention to ensure that multiple viewpoints are represented in the Twitter tree of a topic. For now, the software's special contribution is speed.

“When we have stories that happen very fast and have a very short lifespan, it’s important to have a process that identifies those sources quickly,” Cunningham said.

This story was provided by InnovationNewsDaily, a sister site to TechNewsDaily.