Paper Type
Short
Paper Number
PACIS2026-2118
Description
This study examines how protest narratives shift across cumulative recommendation depths in a YouTube network built from Indonesian protest-related seed videos. Using recursively collected watch-next recommendations, we construct cumulative-depth networks and apply Collective Influence Focal Structure Analysis (CI-FSA) to identify structurally influential focal sets. Narratives are labeled using a large language model and grouped into three clusters: Apolitical, Political-Government Response, and Political-Grievances/Protests. Results show that at shallower depths, focal structures are strongly associated with grievance and protest narratives. As depth increases, protest narratives decline, while apolitical and government-response content becomes more prominent. In contrast, the broader network shifts more gradually toward apolitical content. Government-response framing remains dominant within focal structures across depths. These findings suggest that structurally salient regions of the observed recommendation network exhibit depth-sensitive patterns of narrative redistribution, showing how protest-related, government-response, and apolitical narratives become differently distributed as users move deeper into platform-mediated recommendation pathways.
Recommended Citation
Agarwal, Nitin and Bhuiyan, Md Monoarul Islam, "Narrative Shifts in YouTube Recommendation Networks: A Depth-Based Analysis of the Indonesian Protest" (2026). PACIS 2026 Proceedings. 20.
https://aisel.aisnet.org/pacis2026/ai_ethic/ai_ethic/20
Narrative Shifts in YouTube Recommendation Networks: A Depth-Based Analysis of the Indonesian Protest
This study examines how protest narratives shift across cumulative recommendation depths in a YouTube network built from Indonesian protest-related seed videos. Using recursively collected watch-next recommendations, we construct cumulative-depth networks and apply Collective Influence Focal Structure Analysis (CI-FSA) to identify structurally influential focal sets. Narratives are labeled using a large language model and grouped into three clusters: Apolitical, Political-Government Response, and Political-Grievances/Protests. Results show that at shallower depths, focal structures are strongly associated with grievance and protest narratives. As depth increases, protest narratives decline, while apolitical and government-response content becomes more prominent. In contrast, the broader network shifts more gradually toward apolitical content. Government-response framing remains dominant within focal structures across depths. These findings suggest that structurally salient regions of the observed recommendation network exhibit depth-sensitive patterns of narrative redistribution, showing how protest-related, government-response, and apolitical narratives become differently distributed as users move deeper into platform-mediated recommendation pathways.
Comments
03-EthicsSocietalImpact