Embedding Regression for Comparative Discourse Analysis: Tracing Far Right Ideology Across Media Ecosystems

This study examines the ideological evolution and convergence in political communication across two of the most influential mainstream conservative and far-right media platforms between 2005 and 2020. It analyzes discursive differences within and between platforms using a novel computational approach that combines a la carte (ACL) word embeddings with embedding regression. Changes in racial and political discourse are measured using data from a large corpus of broadcast news transcripts and online forums. The results reveal an uneven landscape of mainstream radicalization but indicate a significant convergence between mainstream and white nationalist discourse toward the nation and immigration. This study contributes a novel methodological approach to the quantification of political discourse across multiple platforms. Substantively, it extends our understanding about the ways fringe ideologies can diffuse into mainstream political dialogue. Finally, this paper contributes to the scientific study of the evolving nature of conservative media ecosystems in the digital age.