Going into last week’s Britain’s Got Talent (BGT) show, the favourites were Calum Scott and Côr Glanaethwy. We here at FiO always thought that Jules O’Dwyer and her dogs had more of a chance and this is the reason why:
The key to predicting who will win Britain’s Got Talent is not so much observing the most talented people or groups but analysing the people who are voting. With Elections there are opinion polls, however for something like BGT, which is influenced by social drivers; social media is a good place to look.
We used a social media listening tool to extract over 90,000 tweets for the 10 finalists over the whole course of the show (1st April to 31st May) but excluding the 2 wildcard acts. We then performed sentiment analysis on these tweets using Python.
Key indicators of popularity
For our analysis, we were mainly looking at 3 key factors to help us decide who would win. These factors were: –

  1. Volume of twitter mentions
  2. Sentiment or how positive the tweets were
  3. Twitter volume growth – how did the volume of coverage change over the course of the competition.

Why were these factors considered?
Volume of twitter mentions is an obvious indicator of the impact that an act makes. We looked at 2 periods; from the start of the show up until a day before the finals, and secondly after the final performances but before the results were announced. It is important to remember that the volume for a particular act maybe high but not always for the right reasons! Hence, sentiment was an important consideration.
Sentiment scale is between 1 and -1with the more positive a mention is the higher the sentiment score and vice versa. It’s expected that the more positive the public feel about a particular act, the more likely it is that they would make the effort to pick up the phone and vote.
However we found that a further factor was necessary to differentiate between the acts – these were all talented acts, so understanding the momentum, or how public perception was growing was also important.
Sentiment analytics
Just before we give you the results of the model, it may be worth going into a bit of detail around what sentiment analytics is and how it works. We illustrate this with some real tweets: –


Sentiment analysis aims to determine the attitudes/emotions of the speaker. The analysis is based on a predetermined list of positive/negative words with different weightings applied to different words. So for example, the word ‘love’ would have a higher positive score than the word ‘like’. Taking a look at the first tweet, positive words such as ‘win’, ‘love’ and ‘beautiful’ are used and this gets a high positive sentiment of 0.72. The second tweet is just a statement which doesn’t use any positive or negative descriptive words and so is neutral and the same logic applies to the third.
Model results
By combining all of the above factors, we can start to build a fairly powerful model of who could win.
How well did this model perform though?
BGT Winner predictions after the Final performance:

Volume growth in tweets from semi-finals to finals, would indicate that the acts final performance grabbed viewers’ attentions, while the higher the sentiment, the better the public perception hence any act in the top right quadrant has a chance of winning. This however is not the only consideration as we also need to consider the size of the bubble. Having a high momentum and sentiment but low volume would most likely mean that that contestant only has a small fan base and thus wouldn’t pull in enough votes to win.
The actual winner of BGT was in fact what the model predicted:
1st = Jules and Matisse
2nd = Jamie Raven
3rd = Côr Glanaethwy
So the model worked for the final – but did it work so well in the semi-final?
BGT Winner prediction at semi finals (or before finals):

Looking at the winner predicted before the final performances took place, The model predicted:
1st = Côr Glanaethwy
2nd = Jules and Matisse
3rd = The Neales (This act nearly made Simon Cowell cry at the semi-final stage!)
What this shows is the importance of the final performance, and in particular the Twitter volume growth factor. Côr Glanaethwy delivered a good final performance but didn’t win because although they were consistent throughout the competition, Jules & Matisse and Jamie Raven increasingly flourished throughout each stage of the competition. Furthermore, both their final performances were more appealing in terms of playing with the audiences’ emotion; Emotion helps create memory so it comes as no surprise that the winner was a close call between Jules & Matisse and Jamie Raven.
This is only a small insight into what Social Media Analytics can do, however this analysis can be used to support businesses, such as helping to determine which products are most popular with the public so that shops can evaluate their stock strategy. Who knows – perhaps this analysis can eventually replace the opinion pollsters in future elections.