Predicting Elections Results using Social Media Activity A Case Study: USA Presidential Election 2020

Abstract

Utilizing social media to ascertain its user's opinions over an entity, more specifically Twitter to forecast Trends is a popular field of research. Employing Twitter for election campaigns, to monitor predict election results, capturing the say of both voters candidates in real-time has escalated over the years. Users share their views, preferences on Twitter voluntarily and are publicly accessible. 'Tweets' are quick, brief real-time user updates and can be extracted through Twitter API. Downloading replies on the most recent Tweets of the candidates can serve our purpose. Location parameters can help in obtaining more accurate credible results. in this paper, we have compared four methods from machine learning and deep learning domains named as 'textblob', 'naive bayes method 'support vector machine' and 'BERT based deep learning approach' for sentiment analysis. we found that BERT based approach is superior to others. To demonstrate the performance of these methodologies we have taken a recent USA presidential election 2020 as a case study. © 2021 IEEE.