Polarized social media networks: a novel approach to quantify the polarization level of individual users

Abstract

ABSTRACT This paper presents a novel methodology to quantify users’ polarization within social media networks, focusing on network structure instead of content. We model polarized networks as directed graphs, collecting users’ follower/following connections data, and propose a simple algorithm to compute a Connectivity Score for each node in sub-graphs associated with identified contrary groups of the polarized network. Our method has several advantages: it relies solely on users’ follower/following connections data, which is objective, often public and easy to obtain. Also, this input data is usually memory-efficient, unlike the input data of content-based methods which may require a whole corpus. The algorithm's performance is assessed through application to Twitter's 116th Congressmembers network, comparing against a content analysis method: Sentiment score (using VADER), and two political behavior-based methods: Ideology score (based on co-sponsorship frequency) and Roll Call score (based on bill-voting similarity). The results demonstrate that: (1) users’ choice of connections on the social media can represent polarization behavior; (2) a meaningful correlation between Connectivity Scores and Ideology or Roll Call Scores shows that the political behavior of users is reflected in their social media connections; (3) Democrats’ Twitter following behavior and their bill-voting and bill co-sponsorship behavior (represented by Roll Call and Ideology) are all significantly more correlated than that of Republicans. We believe applying our algorithm in conjunction with other methods is a valuable contribution resulting in more comprehensive analysis of the social media polarization space.