The Powerful Machine Learning Behind Bing Predicts
The last time I was in Redmond, I got the opportunity to speak with the Bing Predicts team. The Bing Predicts team uses search, social, and other relevant data to make intelligent predictions about upcoming events, like sporting events, reality TV shows, award shows, and political elections.
I first spoke with Walter Sun, the founder and technical lead of Bing Predicts. In How It Began , he shared how he first got the idea of using Bing’s search data along with other relevant data sources to make predictions. They started small and then expanded into many categories of predictions.
Each of these categories of predictions presents its own unique challenges. They began with Reality TV , predicting who would go home each week on shows like American Idol, The Voice, and Dancing with the Stars. These shows are essentially popularity contests where the public votes for their favorites, so utilizing Bing’s search data as well as data from social media proved very effective.
They also make predictions on the winners of Awards Shows , like the Grammy Awards, Golden Globe Awards, and Academy Awards. These pose different challenges, as these awards are determined by a committee of judges, not the general public. An esteemed veteran of the film industry on that committee may not have the same opinions as the general public and the demographic on Twitter and Facebook.
They also predict Politics & Elections results, like the Scottish Referendum and the United States Presidential Election. Although these are also popularity contests determined by the masses, this has its own unique challenges as well: they need to be able to differentiate if users are searching for “Donald Trump” because they support him or because they want to watch negative portrayals of him. Support and interest aren’t the same thing!
Finally, they make predictions on Sporting Events like the NCAA March Madness tournament, the NBA playoffs, the NFL season, Wimbledon, the Cricket World Cup, the English Premier League, and more. This one is hardest to understand: more users searching for “University of Michigan Wolverines” than “Michigan State Spartans” doesn’t affect how the teams will play. But the Bing Predicts team is still able to use this data to get a 5% boost in their accuracy numbers. (Learn more in the video!)
The Full Series of Bing Predicts videos