Motivation

This project was inspired by the Youtuber NotDavid, who built a similar network to analyze his subscribers for a 100,000 subscriber special. The method of detecting overlap with commenters was largely based on the YoutubeAtlas by Kiran Gershenfeld, which aims to map out the entire Youtube network.

Methods

I first pulled my subscriptions (∼150 seed channels), creating a directed ego network. To create a co-citation network, the product of the adjacency matrix with its own transpose was found. I then used 3 latest videos of each of my subscriptions to find 50 commenters for each channel and pulled their subscriptions as well. To stay within the daily free API limit of 10,000 queries, I made a simple SQLite database that could store data, with separate programs for “harvesting” day by day and building the network. Information including the channel name, subscriber count and video titles were also stored. To reduce computational strain, nodes are reduced from 190,000 to∼2300 by filtering for the degree of the node (the minimum links), while links are reduced from about 400,000 to∼71,000 by filtering for links which have a higher weight (more overlap).

Analysis

The Leiden algorithm for community detection was implemented, and I categorized each community with a few descriptors by reading their top-subscribed channels and video titles. 32 communities were found with resolution parameter 1.3, and the network visualized through the open-source software Gephi. The in-built algorithm ForceAtlas2 was used to spatialize the weighted undirected graph in 2D.

Insights

Interestingly, a distinct asymmetrical feature (turquoise) in the network is observed. Analysis of the channels associated shows that this very segregated community mainly consists of Chinese (Cantonese / Mandarin) channels, showing the bilingual nature of the viewer. This suggests that the internet, while extremely interconnected, is quite separated by language. In the graph, it is observed that the connection between this community and the wider network contains news, music – highly subscribed channels (navy), which also indicates how world news and music transcends language.