The Algorithm vs The ‘Experts’

Text Mining Solutions has now processed over 40 million tweets! Having collected such a large volume of data and already shared a great deal of this with our readers, TMS believes that with the election just a week away, it is time to demonstrate, or indeed reiterate, just how accurate and reliable social listening can be (if you don’t already believe it!).

This week, TMS can reveal a vote share prediction for Scotland, and a subsequent seat model, based on tweet analysis over the course of the election period. By using the various facets of a tweet, such as the text itself, user descriptions and mentions, the algorithm has been able to classify each tweet according to the indicated political preference of the tweeter. It is the detailed and thorough analysis of tweets, and the measures taken to avoid misclassification, that allows Text Mining Solutions to then complete a vote share calculation with accurate results. Using this method, vote share predictions can be completed on various levels, from global, to national, to local. This can even be applied to York, and in the next few days we will be sharing candidate preferences and party performance across inner and outer York.

Below, is a Text Mining Solutions Pie Chart which reveals our predictions for vote share in Scotland.

It demonstrates a clear win for the Sottish National Party, followed by the Tories and then Labour.

From these results, TMS has been able to calculate a seat model prediction. By using the percentages from the vote share and using a calculator created by Electoral Calculus, TMS can reveal the following:


Breaking this down further, we are even able to share a list of the predicted seat changes. Huge news from this is that Jo Swinson could be losing her seat!

So there you have it, the General Election predictions for Scotland, by Text Mining Solutions.

But this week’s first blog post is not complete without putting our algorithm alongside the ‘experts’. In order to demonstrate the reliability of social listening, and indeed the Text Mining Solutions algorithm, the table below places our prediction results for Scotland alongside those of the ‘experts’. These include the Panelbase, Electoral Calculus and YouGov predictions, and are the ‘traditional’ polling companies, so to speak. Whilst the BBC generates predictions based on the averages of all poll companies, these poll companies use the traditional method of telephone research to create their predictions.

As demonstrated below, the Text Mining Solutions predictions compare very well to all of these. The vote share values are very similar for all parties, which suggests that social listening is an accurate and reliable way to collect data regarding political views, which is representative of the population as a whole.