Identifying content traps in recommendation networks: comparison between regular and collective influence-based focal structure analysis

Content traps are structurally cohesive regions in recommendation networks characterized by high attraction and retention. This study compares two methods for identifying such structures, Focal Structure Analysis (FSA) and Collective Influence-Focal Structure Analysis (CI-FSA), to determine which better detects structurally critical content traps in YouTube recommendation networks. We construct recommendation graphs from two discourse domains (2025 U.S. tariff expansion and 2024 Taiwan presidential election), apply both methods to extract focal sets, and evaluate them using trap-intensity scores (measuring attraction and retention), structural properties (size, centrality, clustering, participation coefficient), and network resiliency metrics (connectivity loss, fragmentation). CI-FSA identifies focal sets with 14-fold lower median size than FSA, while exhibiting substantially greater structural impact: 24% higher average trap intensity, 11-fold higher median eigenvector centrality, and twice the connectivity loss per removed node. Structural analysis reveals that CI-FSA selects nodes that bridge multiple communities, whereas FSA extracts larger, community-embedded regions with lower criticality. These findings suggest that small, propagation-critical bottlenecks play an important role in the structural organization of content traps, and that CI-FSA identifies such regions more effectively than FSA. For platform design, these findings suggest that targeted interventions on approximately 20-node structural bottlenecks may be more efficient in disrupting network connectivity than interventions on larger, approximately 300-node community structures.