Monday, May 11, 2015

Comparing Viewer Base of Popular LoL Streamers Using Hierarchical Clustering

Have you wondered how League of Legends streamers are "similar" or "different" by their audience? How does Dyrus's viewer base different from Froggen's?

The following diagram showcases the similarities and differences for some popular LoL streamers based on their viewer data. Streamers that are closer to each other on this tree diagram are far more likely to be watched by the same viewer.

Similar to what I did for Nemesis Draft, I used hierarchical clustering (with the "average" metric, in case you care about the details) to analyze 100 popular streamer using their registered viewers data from April 27th to May 2nd, 2015. I have also outlined 10 major clusters/groups of these streamers.

In case you are curious if your favourite streamer is on the graph, here's the full list.

Some quick observations:

1. There are some very obvious clusters by the language line. It's easy to see that the first box from the top consists of Chinese streamers and the second box consists of Turkish streamers and Riot Games' official Turkish channel. This is unsurprising, since a viewer who understands Turkish is far more likely to watch any of  the Turkish streamers than the Chinese ones.

2. English, non-professional-player streamers (box box, cowsep, nightblu3, wingsofdeaths) have slightly different viewership base from the professional player streamers such as dyrus and wildturtle. European streamers (froggen, cyanide, rekkles) also capture slightly different audience from the rest of the English streamers.

3. The last cluster consist of nicktron, kneecoleslaw, kaceytron, destiny et al.) are interesting. I have not watched all of these streams - but to my best understanding many of these are not very high elo players; however, they are entertaining by their own ways.

Why should you care about this graph:

1. As a viewer, you can use this graph as a reference to find your next favourite stream. For example, according to this diagram, Doublelift and Sneaky have similar audiences. So, if you enjoy Doublelift's stream, maybe you should also give Sneaky a shot - there is ample amount of statistical evidence that supports this.

2. As a streamer, you can use this kinds of graphs to gauge your audience. For example, if you are Doublelift and you know your audience is similar to Sneaky's, it might be interesting to stream in different hours, or start streaming right around Sneaky shutting down. While I am sure Doublelift does not need any additional viewers, this may be helpful for smaller streamers who are just starting to stream more seriously.

I would like to sincerely thank @brettfarrow who taught me how to capture viewer data via Twitch API. This analysis would not be possible without his assistance. If you like Twitch statistics you should definitely follow him on Twitter.

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.