Can social media reveal the preferences of voters? A comparison between sentiment analysis and traditional opinion polls

Abstract

This study aims to determine whether the results obtained by applying sentiment analysis to data extracted from social media can reveal the political preferences of citizens to a greater degree of accuracy than traditional public opinion surveys. The researcher collected and analyzed 92,441 tweets related to the Brazilian presidential candidates during the second round of elections in 2014. The analysis results were compared with six voter preference polls conducted by the Datafolha Research Institute. The results showed that sentiment analysis may indicate voter preferences at an accuracy that was only 1% to 8% different than that of traditional research, which has an average accuracy of 81.05%.