Analyzing Political Polarization in News Media with Natural Language Processing

Abstract

Political discourse in the United States is becoming increasingly polarized. A Natural Language Processing (NLP) framework is provided to uncover political polarization, utilizing political blogs and political websites as unique sources of insight into this phenomenon. These aspects of polarization are quantified utilizing different machine learning classifiers and methods of analysis. The creation of an ensemble learner as a voting system is proposed for increased accuracy of text classification of documents between differing political poles. Methods are applied to study polarization across eleven differing political sources from the timeline of January 2020 to July 2020. Findings indicate that discussion of events across this timeline are highly polarized politically, and a list of the top identifying features (words) which may indicate significant slant is provided. This work contributes to a deeper understanding of computational methods for studying polarization in political news media and how differing political groups manifest in language.