Recommender Systems and Democracy: Linguistic Diversity, Epistemic Agency, And Civic Participation

Abstract

The thesis examines how recommender systems (RS) affect democracy by reshaping linguistic diversity and epistemic agency. The objectives were to analyse how RS shape the language and perspectives users encounter, to compare policy approaches to AI in different jurisdictions, and to examine how RS are framed in research discourse. The study employed qualitative and comparative methods across three strands. Case studies of YouTube and Dzen used sock puppet accounts to capture recommendation outputs, which were analysed through critical discourse analysis to trace changes in linguistic range and sentiment distribution. A comparative analysis examined AI policy frameworks in the EU, US, and Russia through official documents and strategies. A third strand analysed research publications on RS, mapping problem framings, proposed solutions, and the treatment of social and linguistic issues. The analysis shows that RS narrow linguistic diversity by amplifying dominant discourses, reducing interpretive possibilities and epistemic agency. Governance approaches diverge: the EU emphasises risks and rights, the US prioritises markets and innovation, and Russia focuses on control and information management. These differences reveal the absence of a common democratic framework. Research discourse is dominated by technical and solutionist framings that marginalise social and linguistic concerns. The thesis contributes by providing evidence of linguistic effects of RS in English and Russian, showing similar patterns of omission across democratic and authoritarian contexts, and demonstrating how discourse separates researchers, policymakers, and users. It positions RS as cultural technologies of omission, placing them within a broader lineage of systems that narrow and filter possibilities, and highlights their implications for democratic processes.