You are what you browse: A robust framework for uncovering political ideology
share
The political opinion landscape, in a democratic country, lays the foundation for the policies that are enacted, and the political actions of individuals. As such, a reliable measure of ideology is an important first step in a river of downstream problems, such as; understanding polarization, opinion dynamics modeling, and detecting and intervening in disinformation campaigns. However, ideology detection is an inherently difficult task, and researchers encounter two main hindrances when approaching an ideology pipeline. Firstly, the ground truth that forms the basis for ideology detection is often labor-intensive to collect and becomes irrelevant with time. Furthermore, these sources are often biased and not robust between domains. Secondly, it is not clear through what lens to view users to infer their ideology, given a small set of users where this ideology is known. In this work, we present an end-to-end political ideology pipeline, which includes; a domain-independent ground truth based on the slant of media users' share, a socially-informed lense allowing performant ideology inference, and an appropriate classifier methodology. We apply the pipeline to both the conventional use case of left-right ideology detection, and the detection of far-right users (who are often of more concern). The ideology detection pipeline can be applied directly to investigate communities of interest, and sets a strong footing for a plethora of downstream tasks.
READ FULL TEXT