Exploring Political Polarization in México: Automatic Classification of Comments on YouTube

Abstract

YouTube, the second-largest social network globally, hosts over two and a half billion monthly users, with content surpassing five hundred hours up loaded every minute [1]. Channels dedicated to news and political discourse fa cilitate interactive communication, enabling users to critique, express viewpoints, and protest anonymously. Partisan engagement on social media is highly contro versial and can influence the attitudes and behaviors of individuals and organiza tions opposing views [2]. Amid growing concerns about political polarization in Mexico, the fourth country with the highest number of YouTube users, this study aims to understand digital communication patterns and their impact on user attitudes. Web Scraping and Natural Language Processing techniques were employed to gather and analyze comments from two antagonistic political channels on YouTube: "El Chapucero" and "Atypical TV". The objective was to identify key aspects of polarization in the comments of users of these YouTube channels to create a Machine Learning model capable of predicting a user's political stance. Distinct features in the dataset were highlighted to train four Machine Learning and Deep Learning classifiers: Naive Bayes, Logistic Regression, CNN, and Bi directional LSTM. These classifiers were able to automatically infer the political leanings of users, the one that performed the best was CNN with a precision of 96%. The main contribution of this study lies in the word analysis that provide insights into the Mexican partisan dynamics on YouTube and in the precision of comment classification, which is achieved due to the polarization existing be tween these political opinion channels.