Modeling Polarization on Social Media Posts: A Heuristic Approach Using Media Bias

Sadia Kamal, Jade Gullic, Arunkumar Bagavathi

Abstract: Developing machine learning models to characterize political polarization on online social media is challenging due to the lack of annotated data, noise in social media datasets, and large volume of datasets. The common research practice is to analyze the biased structure of online user communities for a given topic or to qualitatively measure the effects of polarized topics on social media. However, there is a very limited work to analyze polarization at the ground-level like the social media posts itself which are heavily dependent on annotated data. Understanding the level of political leaning in social media posts is important to quantify the bias of online user communities. In this work, we show that current machine learning models can give better performance in predicting political leaning of social media posts. We also propose two heuristics based on news media bias and post content to collect the labeled data for supervised machine learning algorithms. We experiment the proposed heuristics and machine learning approaches to study political leaning on posts collected from two ideologically diverse social media forums: Gab and Twitter without the availability of human-annotated data.