Utilizing NLP for Stance Detection and Analysis of Political Polarization in Social Media

Political polarization occurs when individuals or groups adopt increasingly extreme political beliefs and attitudes, becoming more entrenched in their positions. Traditionally analysed through surveys, studying these attitudes can be costly and time-consuming, particularly when examining polarized subpopulations such as fringe communities.

Recent advances in NLP models offer unprecedented opportunities to study polarization “in the wild,” providing insights into the attitudes of different communities. We propose using NLP to extract attitudes (i.e., stance detection) from individuals’ texts and applying network models to infer the underlying belief systems. By examining changes in beliefs on an individual basis, this project will explore whether belief systems are increasingly diverging (i.e., polarizing) over time and how the initial structure of a belief system influences the polarization process. Additionally, measuring the distribution of attitudes across different topics will provide insights into whether political polarization also extends to non-political areas, such as music and gaming.