Date of Award

5-20-2026

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Information Science

First Advisor

John Talburt

Abstract

Recommendation systems play a major role in shaping how users encounter information on digital platforms, yet their influence often unfolds through sequences of recommended content rather than isolated outputs. This dissertation investigates recommendation bias on YouTube by examining how recommendation pathways evolve across long-form videos and YouTube Shorts. Focusing on politically sensitive narratives, including the China–Uyghur conflict, the South China Sea dispute, and Taiwan’s 2024 presidential election, the dissertation develops a unified framework for understanding how recommendation systems reshape exposure across platform formats. Using an audit-based and multimodal methodology, the dissertation combines controlled recommendation collection, topic and relevance analysis, emotion assessment, moral foundation analysis, engagement analysis, recommendation-network analysis, thumbnail interpretation, symbolic-content assessment, keyframe-based visual auditing, and watch-time-sensitive simulation. The findings show that recommendation bias on YouTube is not best understood as a static ranking problem. Instead, recommendation pathways exhibit cumulative drift away from the original seed narrative across multiple dimensions. In long-form YouTube, this drift appears through changes in topic relevance, emotional tone, moral framing, engagement concentration, and recommendation-network structure. In YouTube Shorts, similar drift is observed in a more visually immediate and behavior-sensitive environment, where recommendation pathways show relevance decline, topic redistribution, engagement shifts, and visual-semantic change through thumbnails, symbolic imagery, and keyframes. Based on these findings, the dissertation advances the concept of multimodal algorithmic drift to explain how recommendation systems reshape exposure across semantic, affective, structural, and visual dimensions. Overall, the dissertation shows that YouTube’s recommendation systems influence how users encounter contested narratives, with implications for fairness, information diversity, public discourse, and platform governance.

Download

Included in

Social Media Commons

Share

COinS