Concurrent Streaming, Viewer Transfers, and Audience Loyalty in a Creator Ecosystem: A Minute-Level Longitudinal Study
Maxwell Shepherd
Abstract
Live streaming platforms host interconnected communities of content creators whose audiences overlap and interact in ways that are poorly understood at fine temporal resolution. We present a descriptive longitudinal study of audience behavior within a creator ecosystem, analyzing 2.9 million minute-by-minute viewership observations across 7,762 livestreams from 18 affiliated channels over 3.3 years. We find that (1) concurrent streaming is associated with substantial raw per-stream audience decreases (14,377 to 6,057 viewers as concurrent stream count rises from 1 to 9), though hour-of-day controls reduce the residualized correlation to , indicating that scheduling confounds account for much of the observed drop; (2) algorithmically detected viewer transfer events achieve a median efficiency of approximately 50\% across 3,243 candidate events; and (3) audience loyalty metrics (stability, competition resistance, retention, and floor ratio) vary substantially across creators within the same organization, with competition resistance ranging from 0.36 to 1.00, indicating that audience exclusivity is a creator-level rather than organization-level property. These findings provide practical benchmarks for creator organizations making scheduling, cross-promotion, and talent management decisions.
AI Impact Assessments
(3 models)Scientific Impact Assessment
1. Core Contribution
This paper presents a descriptive longitudinal study of audience dynamics within the Hololive English VTuber ecosystem, using 2.9 million minute-level viewership observations across 18 channels over 3.3 years. It addresses three questions: (1) whether concurrent streaming among affiliated creators dilutes per-stream audiences, (2) how effective viewer transfer ("raid") events are, and (3) how audience loyalty varies across creators within the same organization.
The main novelty lies in the temporal granularity and longitudinal scope applied to a creator ecosystem context. While individual findings are relatively intuitive (concurrent streaming correlates with lower per-stream viewership; about half of viewers transfer during raids; loyalty varies by creator), the paper provides concrete empirical benchmarks where previously only anecdotal evidence existed. The decomposition showing that hour-of-day scheduling confounds explain most of the raw concurrency-viewership correlation is a useful methodological contribution, as is the four-component loyalty framework.
2. Methodological Rigor
The methodology is generally sound for a descriptive study, with several commendable features:
Strengths:
Weaknesses:
3. Potential Impact
The practical impact is primarily for creator organizations and talent managers making scheduling and cross-promotion decisions. The benchmarks (e.g., ~50% transfer efficiency, 3× variation in competition resistance) are actionable. However, the academic impact is more limited:
4. Timeliness & Relevance
The paper addresses a relevant practical question as creator economies grow, and the live streaming industry continues to mature. The emphasis on within-ecosystem competition versus cooperation is timely given the growth of multi-creator organizations (MCNs, VTuber agencies, streamer collectives). However, the specific focus on a niche VTuber agency limits broad relevance. The 3.3-year temporal span is a genuine strength relative to typical short-window streaming studies.
5. Strengths & Limitations
Key Strengths:
Notable Weaknesses:
Overall Assessment
This is a competent descriptive study with impressive data scale, applied to a niche but growing domain. It provides useful empirical benchmarks and demonstrates methodological care in acknowledging limitations. However, it lacks theoretical depth, causal identification, and external validity. The contribution is primarily empirical-descriptive rather than theoretically generative, and the narrow focus on a single VTuber organization limits its broader scientific impact. It would be well-suited for a workshop or short paper at a venue like CHI or WebSci, but the scientific contribution is incremental relative to a top venue.
Generated Apr 5, 2026
Comparison History (69)
Paper 2 likely has higher scientific impact due to stronger methodological rigor (large-scale, minute-level longitudinal dataset over 3.3 years), clearer quantitative benchmarks, and broader applicability to multiple platforms and adjacent fields (computational social science, media economics, network effects, recommendation/scheduling). Its findings are immediately actionable for creator ecosystems and can support follow-on causal and modeling work. Paper 1 is novel in a timely domain and offers valuable qualitative theory-building, but its small-N interview design and context-specific focus (crypto KOLs) may limit generalizability and downstream reuse compared to Paper 2’s scalable measurements.
Paper 1 introduces a formal theoretical framework and mathematical models (critical mass threshold, hazard models) for population-dependent systems. This approach offers significant theoretical novelty and broader applicability to network dynamics, sociology, and platform economics beyond gaming. In contrast, Paper 2 is a descriptive empirical study with findings that, while practically useful for creator management, are highly specific to the live streaming industry and offer less methodological innovation or cross-disciplinary generalization.
Paper 1 demonstrates superior methodological rigor and scientific novelty by analyzing a massive, minute-level longitudinal dataset to extract quantitative insights into digital ecosystem dynamics. While Paper 2 addresses a critical societal issue (climate vulnerability), it primarily presents a localized GIS web dashboard, which acts more as a practical application of existing data integration techniques rather than a fundamental scientific advancement.
Paper 1 introduces a novel graph-based modeling framework for team sports analysis that captures spatio-temporal complexity, with potential generalizability across multiple sports. Its methodological contribution—modeling attacking plays as directed paths with multi-dimensional information—offers broader scientific impact across sports science, complex systems, and network analysis. Paper 2, while methodologically thorough with a large dataset, is primarily descriptive and domain-specific to live streaming platforms, offering practical benchmarks but limited theoretical advancement or cross-field applicability.
Paper 2 has higher likely scientific impact due to its scale (2.9M minute-level observations over 3.3 years), fine-grained longitudinal design, and quantitatively defined/replicable metrics (transfer detection, loyalty measures) that can generalize to broader platform economics and networked attention research. It offers actionable benchmarks with clear real-world applications for scheduling and cross-promotion, and its methods can be reused across creator ecosystems. Paper 1 addresses an important topic but relies on smaller surveys/interviews, limiting generalizability and methodological leverage compared to Paper 2.
Paper 1 introduces novel theoretical contributions (the 'civic floor hypothesis'), applies cutting-edge computational methods (GPT-4o for zero-shot classification) to cross-national political discourse analysis, and addresses timely questions about political violence and moral reasoning across institutional contexts. It has broader interdisciplinary impact spanning political science, NLP, computational social science, and moral psychology. Paper 2, while methodologically sound with rich data, is more narrowly focused on live-streaming audience behavior with primarily practical/industry implications rather than broad scientific contributions.
Paper 2 is more likely to have higher scientific impact due to a novel, general-purpose methodological contribution: a new centrality notion for stochastic networks via absorbing Markov chains, plus robustness analysis under kernel uncertainty. This is broadly applicable across network science domains (transportation, epidemiology, infrastructure, social and communication networks) and is timely given uncertainty-aware modeling. Paper 1 provides valuable large-scale descriptive benchmarks for a specific streaming ecosystem, but its contribution is more domain-specific and less methodologically generalizable, limiting cross-field impact.
Paper 2 addresses a highly critical and timely societal issue—opinion manipulation in social networks. Its findings on dynamic versus static intervention strategies offer broad implications across network science, political science, sociology, and AI. While Paper 1 provides valuable empirical insights for the specific domain of content creation and live streaming, Paper 2's focus on mitigating or understanding systemic vulnerabilities in public discourse gives it a wider, cross-disciplinary potential for scientific and real-world impact.
Paper 1 likely has higher scientific impact: it synthesizes a fast-moving, cross-domain ML area (attention-based GNNs/graph transformers) with a new taxonomy, comparative tables, and open issues, making it broadly useful and citable across many fields using graphs (cheminformatics, recommender systems, biology, NLP). Its timeliness and breadth are high. Paper 2 is rigorous and valuable but more domain-specific (creator ecosystems) and primarily descriptive, limiting cross-field methodological influence and likely citation reach compared to a comprehensive ML survey.
Paper 1 addresses a more fundamental and broadly applicable problem in network science—controlled node insertion in incomplete networks—with a novel variational framework (AGN) that fills a gap between link prediction and full-graph generation. Its methodological contribution has potential applications across many domains (social networks, biology, infrastructure). Paper 2, while providing useful empirical benchmarks for live streaming ecosystems, is more descriptive and narrow in scope, limited to a specific platform/creator ecosystem with fewer generalizable methodological advances.