Microsoft researchers predict the likely outcome of Eurovision but don’t bet on the UK
Denmark is the most likely country to emerge victorious from Saturday’s Eurovision Song Contest final in Sweden on Saturday, according to a big data prediction model designed by David Rothschild, an economist at Microsoft Research.
The model takes into account various data sources, including prediction markets, polls, historical trends and social media information and has previously made accurate predictions for the 2012 US Election and the 2011, 2012 and 2013 Academy Awards.
Rothschild predicted on Monday that “Only Teardrops” by Emmelie de Forest had a 41 percent chance of winning the contest, a figure that has increased to 54.1 percent now that both semi-finals have concluded. The next most likely winners are Norway, which has a 10.4 percent chance of winning and Ukraine with 6.2 percent.
Eurovision big data 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,” said Rothschild, adding that the U’s entry, sung by Bonnie Tyler has a measly 0.8 percent chance of winning.
He told TechWeekEurope that he has never actually seen Eurovision and that his interest is an experiment to see how accurately the model might be able to answer more serious economic questions and make better business decisions.
He did admit though that a number of his colleagues were “very excited” and that he hoped that this particular application of his model would increase interest in big data.
Rothschild says that his challenge is to convert raw data into accurate predictions using a variety of sources. Prediction markets, including bookmaker odds, are the most efficient sources of predictions while historical markets show trends over time.
Variety of sources
However markets in which buyers trade contracts on the likelihood of an event happening, much like the stock market, are believed to be more accurate as bookmaker markets are one-sided, with the experts setting the odds with the aim of making a profit.
Social media data is also used, however this is used nearer the time of an event when traffic increases, and is used more to ascertain interest rather than solid predictions.
For example, the songs are available on YouTube, where the most popular entrant was Montenegro, but that country was eliminated in the semi-finals. Rothschild suggests that this was due to its “racy” nature [we haven’t checked a translation yet, but there are ninjas in the music video] and was relieved that the next two most popular videos were Denmark and Ukraine.
Other data sources include traditional polling and historical data such as past voting records and expenditure. By using so many different types of information, the system helps to eliminate any systematic bias for any particular source.
The most important sets of data for Eurovision are prediction markets with very little polling carried out, he said.But he does admit that by relying on betting odds and releasing his predictions, there is a danger that there could be a ‘snowball effect’ and they become a self-fulfilling prophecy.
After all, Sweden was the favourite last year and it eventually emerged victorious in Baku.
“Market prices are not perfect, but they’re better than the next-best alternative,” he said, explaining that his team have conducted experiments to determine the potential effect of predictions on later predictions, or results, but concludes that any impact would occur regardless of whether the forecasts were published.
“We build these techniques to answer serious questions. We want to be able to help make a more efficient world by learning more about the business decisions people make and the economic indicators.”
Rothschild will be watching Eurovision for the first time on Saturday, hoping that a country with a high percentage will win – not the UK then. The predictions will update during the event on his Predictwise website.
What do you know about European tech history? Find out with our quiz!