[Paper Review] Collaborations on YouTube: From Unsupervised Detection to the Impact on Video and Channel Popularity

Christian Koch, Moritz Lode+3 more|arXiv (Cornell University)|May 1, 2018

Hate Speech and Cyberbullying Detection11 references3 citations

TL;DR

This paper proposes CATANA, a deep learning-based framework for unsupervised detection of collaborations among YouTubers using face recognition and clustering on user-generated video content. It analyzes a 3-month dataset of 7,942 channels and finds that collaborations significantly boost video views and subscriber growth—often exceeding 100% compared to non-collaborative content—especially for low- and mid-tier creators (10k–1M subscribers).

ABSTRACT

YouTube is one of the most popular platforms for streaming of user-generated video. Nowadays, professional YouTubers are organized in so called multi-channel networks (MCNs). These networks offer services such as brand deals, equipment, and strategic advice in exchange for a share of the YouTubers' revenue. A major strategy to gain more subscribers and, hence, revenue is collaborating with other YouTubers. Yet, collaborations on YouTube have not been studied in a detailed quantitative manner. This paper aims to close this gap with the following contributions. First, we collect a YouTube dataset covering video statistics over three months for 7,942 channels. Second, we design a framework for collaboration detection given a previously unknown number of persons featuring in YouTube videos. We denote this framework for the analysis of collaborations in YouTube videos using a Deep Neural Network (DNN) based approach as CATANA. Third, we analyze about 2.4 years of video content and use CATANA to answer research questions providing guidance for YouTubers and MCNs for efficient collaboration strategies. Thereby, we focus on (i) collaboration frequency and partner selectivity, (ii) the influence of MCNs on channel collaborations, (iii) collaborating channel types, and (iv) the impact of collaborations on video and channel popularity. Our results show that collaborations are in many cases significantly beneficial in terms of viewers and newly attracted subscribers for both collaborating channels, showing often more than 100% popularity growth compared with non-collaboration videos.

Motivation & Objective

• To address the lack of quantitative analysis on YouTube collaborations among user-generated content creators.
• To develop an unsupervised framework for detecting collaborations without prior knowledge of collaborating YouTubers.
• To analyze the impact of collaborations on video and channel popularity, particularly across different popularity classes and content categories.
• To evaluate the role of multi-channel networks (MCNs) in shaping collaboration patterns.
• To provide data-driven guidance for YouTubers and MCNs on effective collaboration strategies for maximizing visibility and revenue.

Proposed method

• CATANA uses a deep neural network (DNN) for face detection and embedding extraction from YouTube video frames.
• It applies clustering techniques to group detected faces into distinct YouTuber identities despite unknown numbers of participants.
• The system identifies channel owners and guests by analyzing face co-occurrence patterns and temporal consistency across videos.
• Collaborations are inferred when two or more YouTubers co-appear in the same video, with outlier filtering to exclude incidental appearances.
• The framework constructs a collaboration graph over a 3-month dataset of 7,942 YouTube channels, enabling large-scale analysis.
• It integrates video metadata, popularity metrics (views, subscribers), and channel categorization for comparative analysis.

Experimental results

Research questions

• RQ1 How frequently do YouTubers collaborate, and what is the level of partner selectivity in collaboration choices?
• RQ2 How do multi-channel networks (MCNs) influence collaboration patterns among affiliated YouTubers?
• RQ3 Which types of YouTube channels (by content category and popularity class) are most likely to collaborate?
• RQ4 What is the impact of collaborations on video views and channel subscriber growth compared to non-collaborative content?
• RQ5 How does the popularity class of collaborating channels affect the magnitude of growth in views and subscribers?

Key findings

• Out of 7,942 channels in the dataset, 1,599 (20.1%) engaged in collaborations, with an average of 2.8 collaborations per channel.
• Collaborations led to a mean popularity growth of 19% to 51% in video views, with 99% confidence, compared to non-collaborative videos.
• Low- and mid-tier YouTubers (10k–1M subscribers, classes 1 and 2) experienced the most significant benefits, with up to 100%+ growth in views and new subscribers.
• Collaborations between creators in the same content category (especially Entertainment and People & Blogs) showed the strongest positive impact on viewership and subscriber acquisition.
• MCN-affiliated YouTubers were more likely to collaborate with others in the same network, suggesting network-driven collaboration strategies.
• For high-popularity channels (class 4, >1M subscribers), collaborations showed little to no measurable impact on growth, indicating diminishing returns at scale.