Computer scientists at Carnegie Mellon University say plowing through millions of tweets to gauge fan sentiment probably isn't going to help you make a mint betting on NFL games, though doing this themselves has given them newfound respect for sports bookies.
The researchers, who will report the findings of a study of three years’ worth of Twitter posts at an upcoming analytics conference in Prague, were trying to figure out if crunching numbers based on the microblogging site’s content could make predicting football game outcomes any easier.
Their conclusion was that using machine learning tools to analyze the tweets (42 million a day during the 2012 season) wasn’t helpful with winning over/under bets or predicting straight game results, but was slightly helpful in picking against the spread (55% accurate). Why? Because bookmakers take into account things like fan sentiment in setting point spreads, and the CMU researchers were analyzing tweets based on volumes of hashtags about specific teams, plus looking for positive or negative words in such posts.
“One thing that surprised us is how hard setting the point spread is to do well,” said Christopher Dyer, assistant professor in CMU’s Language Technologies Institute, in a statement. “And the sports books are very, very good.”
It wouldn’t surprise the researchers if sports bookies are doing a little Twitter analyzation of their own.
However, they probably aren’t getting funding for their research like CMU, whose work on this has been backed by
the National Science Foundation and Sandia National Laboratories. Researchers in recent years have analyzed tweets for various reasons, from predicting elections to spotting disease hotspots.
Math and computing whizzes increasingly have been using their minds and machines to forecast sports results, too, including for major league baseball and college basketball.