The livestream network: using graph theory to analyze shared viewership on Twitch.tv
As more and more viewers turn away from network television, alternative new forms of entertainment and content have entertained the mainstream. Nowhere is this more apparent than Amazon’s Twitch.tv, which they acquired for almost 1 billion dollars in 2015 and is now valued at almost 15 billion dollars. For any unfamiliar, on Twitch streamers and organizations can “go live” and begin livestreaming content, and viewers can watch this content for free. If they choose to subscribe to a channel (at a five dollar fee or free with Amazon prime) they do not have to watch any ads on that particular streamer’s channel. Twitch, while originally intended to be a platform for gaming content, has made waves in the mainstream recently. Ludwig Ahgren’s subathon drew plenty of attention, where Twitch streamer Ludwig went live on the platform for 30 days straight, and livestreamed his entire life for a month in a Truman Show-esque stream. And of course, last election season congresswoman Alexandria Ocasio-Cortez took to Twitch to play popular game Among Us with some high-profile content creators in order to drive youth turnout. Twitch, like most entertainment nowadays, has become incredibly saturated with content, with thousands of content creators for viewers to watch for free.
A reddit user attempted to make sense of this saturation by displaying the Twitch landscape as the graph below (better resolution here: https://i.redd.it/9egwnv2imdf71.png). Each node represents a streamer, with a larger node representing a larger audience; and an edge between two nodes represents some shared viewership between the two streamers, with a thicker node meaning a stronger shared viewership. They have also attempted to cluster the data, and the color of each node represents the category of streamer. An attached spreadsheet provides a legend for each category/color, the list of every node and its size, and the list of every edge and its weight.
Relevant graph theory from the course is on display in this graph. For one thing, while there are no binary strong/weak ties, we can see that two streamers who share a strong mutual viewership with a third streamer are highly likely to have a shared viewership themselves. This is most apparent in the pink cluster in the left-central area (streamers who are in or are related to content organization OfflineTV; nodes to look for: Pokimane, ludwig, Sykkuno), where each of these nodes are extremely bright pink. The accompanying spreadsheet also contains data for weights of all 20 thousand edges in the graph, and one can check that nodes of the highest weights, i.e. the highest shared viewership, form triads within their clusters. It is also interesting to note the clustering, with the giant component in the middle and the smaller components on the edges. Looking at the spreadsheet reveals that the clusters on the edges are streamers who do not speak English (ex: orange/bottom right is Spanish speaking streamers, dark green/top right is Portuguese speaking streamers), and the clusters in the middle are primarily English speaking streamers. Interestingly, in the left, lime green section of the giant component it seems to be somewhat, but not fully, disconnected. These streamers do speak English, but they are primarily from the United Kingdom, meaning their shared viewership is relatively high with each other as all of their viewers are primarily in that time zone, but still fall in the giant component as their viewers can understand the North American streamers as well. Within the giant component, zooming in it is apparent that a viewer for one streamer is familiar with everyone else in this component, but they are more likely to be familiar with streamers of the same category.
It is hard to apply social networking principles to this graph as the edges and the nodes are not necessarily linked. An edge only represents shared viewership and is more or less unrelated to the actual social network between two nodes. Of course, if two streamers collaborate together, or on the off-chance are friends in real life, the weight of that edge can be expected to be greater (some of the thickest edges are between famous streamer couples; ludwig and qtcinderella being an example in the spreadsheet. the edge between ludwig and qtcinderella is qtcinderella’s strongest tie in the graph). But, the edge between two people represents the actions of thousands of people instead of just those two people, and thus slightly deviates from the social networking theory we have learned in class.
Sources:
Reddit post: https://www.reddit.com/r/LivestreamFail/comments/oxxrpu/twitch_atlas_vol_6/
Spreadsheet: https://docs.google.com/spreadsheets/d/1PwGE7BbPq5-8ssM99nw52Dgeo_oskZPBXDNhsFIaDhM/edit#gid=0