We Used Social Content To Predict The Eurovision Winner

The buzz on social media lets us understand audience opinion surrounding popular topics. As expected, the 2017 Eurovision Song Contest was a trending event that generated an immense amount of discussion on social media. By cross-correlating the data semantically, we get a deeper understanding of whether conversations are positive or negative, making it possible to form powerful predictions.

Using social media as a listening platform is a great way to get an immediate understanding of audience opinion on a large scale. This is particularly relevant for an event such as Eurovision which is largely based on viewer's votes.

We monitored social media activity during the Eurovision Song Contest 2017 by using STORY to track mentions of the artists and country names on Twitter. We used Twitter's GNIP firehose to guarantee that we captured every Tweet, irrespective of traffic volumes (at peak times we measured over 250 Tweets/sec). This enabled us to give a statistically accurate account of social communication patterns.

As one would expect, there were well-distinguished peaks of activity at the time of each country's performance. The graph below shows the Twitter yield for the top 6 and bottom 2 countries from the moment the first candidate started (20:30) until the first results were announced (22:45).

We see that the pattern of the top 6 and the bottom 2 are comparable: a sharp peak around their performance and a quick decay afterwards. Interestingly, the peak sizes of winners and losers are quite comparable. Of the top 6, we can see that the eventual winner Portugal (black), has the highest peak and that it is the only top 6 country whose related chatter continues steadily after their performance. Surprisingly though, also Spain (green), who ended last, is the only other country with significant continuing mentions after their performance (even more than Portugal!).

When looking at the cumulative totals, this pattern is even more pronounced: Portugal stands out as the faster riser among the top countries, but Spain overtakes it before the end of the show and continues to rise fastest of all.

So what's going on here? Are social volumes not a measure of popularity? We wanted a better understanding of what people were actually saying about Portugal and Spain so we set up a word cloud of all Tweets mentioning Spain or their artist over a set timeframe.

*Words from various languages with identical or similar meanings were merged and non-relevant terms such as you', is', but', will' etc were filtered out.

The word cloud clearly shows that people were sharing their disapproval of the Spanish act on Twitter. Terms such as rooster, waroftheroosters, deceive, ridiculous, please and stop (the latter two often in combination) represent the negative sentiment and discussions people were having about Spain during the show. Gibraltar (tiny, in orange) somehow also made it to the top 14 and when people bring this topic up in the context of a European song festival, it doesn't signify anything good.

By this stage, we had formulated the hypothesis that Spain's enormous social volumes were the result of their negative perception by the European audience. We decided to test this by creating a simple poll cross-correlating its results. First we measured the total number of Tweets mentioning each country or their artist:

Where, again, Spain is seen to trigger a lot of attention and later winner Portugal doesn't stand out yet. We then overlaid this with the constraint to only include Tweets that also mentioned love.

Suddenly Spain's numbers drop to really low levels, that can be expected from a contestant that ends up last in a competition. Portugal is now the clear winner and only Moldova (eventually ending 3rd) is less than a factor of 4 away of it. Similar results were obtained when cross-correlating with other positive terms like best, winner or 12points.

We verified these findings by returning to our original graph for the cumulative totals and now also filtering the data for the term love.

The results are obvious and speak for themselves. There were nowhere near as many people mentioning love for Spain as there were for Portugal. In fact, Portugal's numbers soar above the rest and continued to rise so steeply that we predicted their victory well in advance of the televoting, which stopped at 22:45 GMT.

Social media allow us to track interest and measure popularity of topics. However, as this study shows, the statement that there is no such thing as bad publicity' does not necessarily hold for contests using social media as their voice. Just counting Tweets and looking at volumes is a necessary, yet insufficient condition for understanding what's going on. In today's socially complex environment, media companies need enhanced toolsets with built-in flexibility for detecting, interpreting and verifying trends and opinions, which otherwise would have remained hidden for the untrained human eye.

Related: See how we used social media to predict the results of the Brexit EU Referendum.

By Taco Nieuwenhuis, Chief Technology Officer at never.no

Being the sole Dutch employee, Taco frequently smuggles cheese and stroopwafels from his home country into Oslo and constantly tries to understand Norwegian culture. In his spare time he's obsessed with cats and penguins, theoretical physics and leads a parallel existence as garagerocker and concert-dj. Not always simultaneously though.

LINK:	http://www.never.no/news/we-used-social-media-content-to-predict-the-e...
	See more stories from neverno

Most recent headlines