Artificial intelligence for elections: the case of 2019 Argentina primary and presidential election

TL;DR

This paper presents a machine learning and big-data analytics approach analyzing Twitter data to accurately predict election outcomes, outperforming traditional polls and providing insights into societal trends.

Contribution

The paper introduces a novel method combining machine learning, big-data, and network theory to predict election results from social media data, demonstrating superior accuracy over traditional polling.

Findings

Successfully predicted Argentina 2019 primary election results

Predicted upcoming election outcomes with high accuracy

Provided a reliable social trend analysis tool

Abstract

We use a method based on machine learning, big-data analytics, and network theory to process millions of messages posted in Twitter to predict election outcomes. The model has achieved accurate results in the current Argentina primary presidential election on August 11, 2019 by predicting the large difference win of candidate Alberto Fernandez over president Mauricio Macri; a result that none of the traditional pollsters in that country was able to predict, and has led to a major bond market collapse. We apply the model to the upcoming Argentina presidential election on October 27, 2019 yielding the following results: Fernandez 47.5%, Macri 30.9% and third party 21.6%. Our method improves over traditional polling methods which are based on direct interactions with small number of individuals that are plagued by ever declining response rates, currently falling in the low single digits. They provide a reliable polling method that can be applied not only to predict elections but to discover any trend in society, for instance, what people think about climate change, politics or education.