Twitter user multiclass classification during US 2020 electoral campaign

Abstract

Due to the unprecedented rise of data content on social media over the last decade, an opportunity for data-based analysis has become a norm in the modern world. Implementing Machine Learning algorithms and Data Science methods virtually every industry changed. One of the most active researching areas in Machine Learning today is Natural Language Processing (NLP), a field of Artificial Intelligence (AI) that allows computers to read, understand, and deduce meaning from human languages. In this paper we applied Natural Language Processing methods and algorithms on two Twitter datasets collected during the US 2020 elections in order to group both users and tweets in multiple categories based on their support for the candidate. The purpose of this work was to establish the possibility to correctly classify these individuals and their individual tweets based on their aggregated opinions and to create a predictive classification model focusing on text analysis. As a result, we constructed, trained and tested multiple models that can help predict the probability of the user’s sentiment toward the candidates based on their tweets. We showed that in 63 % of the cases, we can present high probability of a user’s sentiment classification, according to the amalgamation of their tweets.