Sort by Categories & Channels by Relevancy Instead of Current Viewers
When browsing categories and channels on Twitch the results are sorted by the current viewer count. While this is one way to sort things, it doesn’t facilitate connecting viewers with the content they are most likely to watch. When looking at other content platforms such as Youtube, Netflix, Amazon Prime Video, etc… none of them present viewers with content based solely on how many people are currently watching it but instead use algorithms to present them with recommendations based on the viewer’s previous behavior and the viewing behavior of other viewers. I believe Twitch should have all of the data necessary to sort categories and channels by relevancy to the current viewer where the current view count is one of many factors used.
The current Twitch “front page” has “Live channels” and “Categories” “recommended for you” so clearly there is already a recommendation/relevancy system in place. As a first step using this existing system to place all recommended channels and categories at the top of the list before reverting back to current viewer count sorting would be a big improvement.
With this change, it will then become incredibly important that the recommendations are relevant. To improve the recommendation algorithm, I recommend that the organic path to discovering new channels is heavily weighted. The most common way a channel is discovered in my experience is when the channel is raided or hosted by a channel the viewer already often watches. This pattern can be repeated recursively with each step removed having less weight on relevancy then the one before it. While the “people who watch X also watch Y” type weighting has merit, Twitch is someone unique because of its emphasis on communities. The actions taken between streamers therefore should have at least as much weight as simple common viewership.
Sort Categories & Channels by Relevancy
Use the “front page” “Live channels” and “Categories” “recommended for you” at first
Fallback on concurrent viewer count when recommendations are exhausted
Improve the recommendation system to produce more relevant results