Whether they expect their country to win or they just want to have a fun evening with friends in front of the TV, each year, an estimated 125 million people watch the Eurovision song contest. Each country in Europe has one entry into this singing extravaganza, with participating countries voting, by country, for the winners. This is the event that launched the careers of Abba (Sweden 1974) – who just opened an interactive museum in Stockholm by the way – Celine Dion (Switzerland 1988), and Julio Iglesias (Spain 1970).
If you’re betting with your friends on this year’s winners, and who tomorrow’s biggest stars might be, you might wish you could see into the future. Well, I wouldn’t say I can go that far, but with the help of technology I can certainly help you out with some smart predictions. These aren’t based on a crystal ball – these guesses are based on sophisticated analysis of massive data sets – too big for the human mind to really appreciate.
So who are my hot tips for this year’s contest? I have Denmark, with Emmelie de Forest singing “Only Teardrops” as the most likely winner at 37 percent in my real-time updating list of likely winners. Ukraine’s Zlata Ognevich, at 15 percent, and Norway’s Margaret Berger, at 11 percent, round out the top three most likely winners.
Likelihood to win Eurovision 2013
|OIContestantIO ||E David'sE |
|OIEContestantOIE || EDavid'sE |
|OIEContestantOIE ||E David'sE |
|IEIIIContestantIIIIE || EDavid'sE |
| Denmark ||41% ||San Marino ||2% || Montenegro ||0.3% ||Cyprus ||0.2% |
| Ukraine ||11% || Finland ||1% || Armenia ||0.2% || Albania ||0.1% |
| Norway ||11% || Great Britain ||1% || Austria ||0.2% || Lithuania ||0.1% |
| Russia ||6% ||Rep. Of Ireland ||1% ||Switzerland ||0.2% || Estonia ||0.1% |
|Azerbaijan || 5% || Greece ||1% || Spain ||0.2% || Croatia ||0.1% |
| Italy ||4% || France ||0.6% || Slovenia ||0.2% || Israel ||0.1% |
| Georgia ||3% || Belarus ||0.5% || Romania ||0.2% ||FYR Macedonia ||0.1% |
| Germany ||3% || Moldova ||0.5% ||Belgium ||0.2% ||Iceland ||0.1% |
| Netherlands ||2% || Malta ||0.5% ||Bulgaria ||0.2% ||Latvia ||0.1% |
|Sweden ||2% || Serbia ||0.4% || Hungary ||0.2% || || |
Sources: SkyBet, BetVictor, PaddyPower, 188Bet, Betfair, Bet365, Coral, 888Sport, BoyleSports, 32Red, WBX, LadBrokes, BetFred, BetWay, BWin, SportingBet, Betdaq, YouWin, StanJames, WilliamHill, OddsChecker
To create these likelihoods I consulted many different data sources, the main type being prediction markets: speculative markets created for the purpose of making predictions. To put it simply, users in prediction markets trade contracts on upcoming events. If the event occurs, the contract is worth $1 and if it does not occur it is worth $0. The price that people are willing to trade that contract is collective wisdom of the probability of the event occurring. If it is trading at $0.05, people do not think it will happen, but if it is trading at $0.95 people are nearly certain it will happen. My model collects data from many prediction markets to produce accurate predictions – as shown with my previous work on the general US election and the 2011, 2012, and 2013 Oscars.
Prediction markets work particularly well for Eurovision, firstly, because the markets are extremely robust, and secondly, because the collective wisdom of the users includes the unique politics of the Eurovision voting blocs. This means that an American like me – with detailed knowledge of markets, but little domain knowledge of Eurovision – can make meaningful predictions. What I have to do is to convert the raw prediction market data into a set of probabilities in a method that best ensures the forecasts are accurate. And because I’m aggregating many sources, it helps eliminate any systematic bias for any particular source.
I will also be following social media data, but I expect that data to peak in relevancy much closer to the event. Major events have 2-3 magnitudes (i.e., 100x to 1,000x) more traffic in the immediate lead-up and during the event than before it. Specifically, I will be following the interest levels and sentiment from Twitter and search.
I’ll also be looking at YouTube, which offers also some valuable insight from the millions of video views. It is intriguing that the most viewed entry is one that I see as having a negligible likelihood of winning: Montenegro’s racy video, Who See singing Igranka, has over 1.3 million views. More assuring, the second and third most watched videos are my first and second most likely winners: Denmark (900,000 views) and Ukraine (1.2 million views). With just one exception, the entire top 10 most likely winners are all within the top 15 most watched videos.
As always with my forecast, all forecasts are set to update in real-time during the event. I expect to see some serious movement during the events as the different countries perform. With high expectations on Denmark, Ukraine, and Norway, their performances are going to be especially interesting.
But for me, the most interesting part of this experiment is not the actual result of Eurovision, but the ways we can learn from it and other major events to make more scalable forecasting and predict the future more accurately. It’s the science of applying the same type of effort, the same type of skills, and the same technology across different domains – whether it’s politics, the Oscars or indeed Eurovision.
Continuing exploration of these challenges will help us create the technology to effortlessly form new predictions around new domains, such as business or economic questions, and ultimately help people make better and more efficient decisions.
At Microsoft we are committed to driving the innovations of the future, and these kind of machine learning and prediction engines are a core part of that effort. Check out our Microsoft Research blog to find out more about our prediction technology - and be sure to tune in on Saturday 18th May to see if my hot tips for Eurovision hit the target!