Predicting the EU 2014 Election Results in Multiple Countries Using Twitter

During the latest years, the behavior of users in Twitter has been explored for various purposes, one of the most famous being the prediction of election results. Most works so far make their predictions by focusing strictly on Twitter data and are applied on some data after the elections; hence, they are biased towards the actual results. In the current work we have focused on the 2014 European Union Elections for three countries. We monitored political discussions on Twitter and created time-series of the political parties by extracting various features; at the same time, we aggregated opinion polls to serve as our ground-truth. Based on those features, we tried to predict the election results, publishing our predictions for one country before the elections ended. Our approach achieved low error rates, being better than two prediction websites and suggesting that Twitter can be effectively used for this task.