Researchers at Carnegie Mellon University have promising news for gamblers, economists and impatient sports fans: You can predict the results of NFL football games with Twitter.
It’s the latest research in the budding field of Twitter modeling, and as with much in statistics, it sounds a bit like magic. The researchers pulled several million football-related tweets from the Twitter fire hose during the 2010, 2011 and 2012 seasons. They then analyzed and cataloged the sentiment by team, ran the data set through a number of statistical models, and came upon several that either matched or beat traditional forecasts.
Conventional, non-Twitter prediction methods predict the winning team around 58
percent of the time. But by combining conventional methods with Twitter-based
models, the researchers were able to predict the game winner with 65.9 percent
accuracy. Models that used both Twitter and traditional data also made the most
accurate predictions on other sports-betting metrics, like the combined number
of points both teams scored.
“We find that simple features of Twitter data can match or exceed the
performance of the game statistical features more traditionally used for these
tasks,” the researchers conclude. “It is hoped that our approach and dataset may
be useful for those who want to use social media to study markets, in sports
betting and beyond.”
That, ultimately, might be the coolest takeaway from this research — the
implication that economists could keep plumbing Twitter for insight on a
yet-unimagined range of topics and fields. Twitter modeling isn’t exactly new,
of course: We’ve already seen the network predict elections, the stock market,
box office revenues and the spread of contagious disease. But there’s a
suggestion here that we’re just beginning to unlock Twitter’s predictive
potential: The network could predict any number of real-life phenomena — from
whether your home team wins its next big game to when hit-and-runs will occur.
In an opinion piece for the Washington Post Sunday, sociologist Fabio Rojas
(who has done a bit of work in this field himself) concluded that social
media modeling will be the death of the political polling industry — it’s far
more accurate, he argues, to analyze tweets than poll results.
That augurs an intriguing new world where social media-fueled predictions are
both more common and accurate. Thanks to Twitter, we could someday guess any
number of details about the future ahead of time — down to the results of
football games before they’re even played.
Article By: Caitlin Dewey
Source: Washington Post
Conventional, non-Twitter prediction methods predict the winning team around 58
percent of the time. But by combining conventional methods with Twitter-based
models, the researchers were able to predict the game winner with 65.9 percent
accuracy. Models that used both Twitter and traditional data also made the most
accurate predictions on other sports-betting metrics, like the combined number
of points both teams scored.
“We find that simple features of Twitter data can match or exceed the
performance of the game statistical features more traditionally used for these
tasks,” the researchers conclude. “It is hoped that our approach and dataset may
be useful for those who want to use social media to study markets, in sports
betting and beyond.”
That, ultimately, might be the coolest takeaway from this research — the
implication that economists could keep plumbing Twitter for insight on a
yet-unimagined range of topics and fields. Twitter modeling isn’t exactly new,
of course: We’ve already seen the network predict elections, the stock market,
box office revenues and the spread of contagious disease. But there’s a
suggestion here that we’re just beginning to unlock Twitter’s predictive
potential: The network could predict any number of real-life phenomena — from
whether your home team wins its next big game to when hit-and-runs will occur.
In an opinion piece for the Washington Post Sunday, sociologist Fabio Rojas
(who has done a bit of work in this field himself) concluded that social
media modeling will be the death of the political polling industry — it’s far
more accurate, he argues, to analyze tweets than poll results.
That augurs an intriguing new world where social media-fueled predictions are
both more common and accurate. Thanks to Twitter, we could someday guess any
number of details about the future ahead of time — down to the results of
football games before they’re even played.
Article By: Caitlin Dewey
Source: Washington Post
RSS Feed