Sentiment analysis of tweets for the 2016 US presidential election

Abstract

Twitter is a popular micro-blogging social media platform. For the 2016 US Presidential election, many people expressed their likes or dislikes for a particular presidential candidate. Our work's aim was to calculate the sentiment expressed by these tweets, and then compare this sentiment with polling data to see how much correlation they share. We used a lexicon and Naive Bayes Machine Learning Algorithm to calculate the sentiment of political tweets collected one-hundred days before the election. We used manually labeled tweets as well as automatically labeled tweets based on hashtag content/topic. Our results suggest that Twitter is becoming a more reliable platform in comparison to previous work. By focusing on tweets 43 days before the election (beginning with the first presidential debate), we found a correlation as high as 94% to polling data using a moving average smoothing technique.