Time: 15:51 (Greece)
We are happy to publish a set of predictions regarding the outcomes of the Euro-Parliament elections 2014 for Greece making use of an approach developed by SocialSensor. Obviously, such results should be read with great caution as they stem from a highly experimental setting and are only the beginning of some very interesting research that will continue.
How we did it?
We monitored the Twitter stream for mentions of the main contesting Greek parties and computed features reflecting the volume and sentiment around them over time (for instance, number of positive mentions for party X). Then, using published poll results, we trained some predictive models using different algorithms (Linear Regression, ε-SVM, Sequential Minimal Optimization, Gaussian Process). For each party, we present the minimum and maximum predicted percentage, together with an average prediction. The results refer to percentages for targeted votes (i.e. not including white/invalid votes).
Here are the results....
Disclaimer: In case it was not clear from the above, those results were produced by means of automatic data analysis methods and they do not reflect any political bias of the researchers. In addition, they should not be considered as reliable vote estimates but rather a research experiment.