The looming cliff of Twitch chat; How to design the UX of streaming chatrooms.

We’re bombarded with streaming services as consumers, but none are utilizing live entertainment as well as Twitch, the live video gaming platform.

With their eSport competitions, music coverage, and crowd-sourced creative content, they’ve become one of the fastest growing platforms in the game. In fact, Twitch reports that an average of 15 million people PER DAY watched or participated in their live streams [Link]. No wonder Amazon bought this company for almost a billion dollars.

But here’s the stitch with Twitch. Many of their core features haven’t scaled to meet the demand for these large, hungry audiences.

The Product Problem:

The UI and user experience of Twitch chat was never meant to be used with 10,000 users simultaneously. Solution, implement a contextual algorithm to determine which comments display to Streamers.

A single theme unites Streamers. The wish to have more viewers, more followers, a larger audience. How to feed this need? There’s also direct positive correlation between the amount of chat activity and a Streamers channel size. Every Streamer realizes their participation in chat bolsters engagement so finding ways to nourish chat activity is always an issue — but how can they chat with thousands of viewers at once?

With way too many people chatting but an increased need to keep conversations going, there might be opportunities for using chat algorithms to help push things forward for Twitch.

Have you noticed those word suggestions popping above your keyboard when typing on your phone? Those are determined by the context of your conversation. As you type your phone’s trying to understand what you’re wanting to say and helps you along. This machine learning/algorithmic design makes you more efficient. Other applications are hopping on this bandwagon too. Gmail offers quick email responses in the form of 3 to 5 word canned auto responses.

Our theorem below can help us craft a chat algorithm of our own.

VE = Viewer Engagement | SE = Streamer Engagement | SS = Number of Streaming Sessions | HUG = Holistic User Growth

Verbally this theorem states that increases in viewer engagement increase streamers’ engagement, these increase number of overall streaming sessions, all which trigger a cycle of platform growth.

If a viewer participates in chat they’re way more likely to subscribe, follow, or donate money to a Streamer (ΔVE). As well, having Streamers converse with viewers (ΔSE), triggers significant chat participation, but this is harder at scale for Streamers who broadcast to large numbers. Some Streamers employ moderators, secondary admins or friends to help with the onslaught of chat messages. Sadly this solution doesn’t mend issues with balancing response rates for those tuning in.

Instead of trying to redesign an entirely new system, finding small ways for improving the user experience can go a long way.

It’s not realistic to think Twitch will design an entirely new chat experience so refining what’s already there is a better goal to set. With the algorithm produced using this theorem and slight UI changes we can deploy a much richer chat experience for everyone. Increasing the number of users who feel recognized by the Streamer will grow that performer’s ability to provide quality entertainment to larger audiences.

A new chat window design for Streamers would follow a set of rules, or algorithmic formulas surfacing specific viewer comments, E.G. first time viewers, and users who don’t often post in chat.

As seen in this concept, some comments maintained prominence in the chat window while others moved out of view in their normal cadence. We can see the comments made by Kore and Sout1337 are enlarged and moved to the top. This helps Streamers and moderators notice the comment. Along with making their comments bigger, it holds certain comments longer in the chat window, pinning it to the top for a set period of time before timing out and removing it from view.

It’s important to understand these UI changes would only be visible to Streamers. Viewers would never see their comment pinned or know it was promoted.

Counterpoints to this design would be that pinned/highlighted comments could be of lower value and quality. Adding variables like Key Word Detection, Type of Twitch User, Previous Follower/Subscriber, or even Verbal Language Detection on the Streamers side could increase the algorithm’s effectiveness — but using only a few variables would do the trick for diversifying Streamer chat activity.

Variables for determining comment promotion:

• Twitch Member Length (new to Twitch or long-time member)
• User Relationship to Streamer (subscriber, follower, Streamer follows that viewer also, donations made)
• Most Recent Comment (within the current stream or in total overall viewing sessions on twitch)
• Number of total global comments across all streaming sessions (not tied to single Streamer)
• Types Of Comments (number of characters, keyword detection, question mark used within the text)

Viewers realize their comments are drowned out by the sheer number of participants, thus why number of comments doesn’t increase linearly 1–1 as viewer count grows. The proposed theorem and designs solve for commenting over saturation. This allows Streamers to engage more deeply, converting passive viewers into engaged audiences. By improving the consumers(viewers) experiences we improve the providers(Streamers) experience.

Implementing this design would be great, but how do we measure success? Number of comments, followers, donations? Or perhaps more difficult qualitative metrics like Streamer and viewer satisfaction? Whichever way, from a business standpoint, the main goal is quality platform growth. Twitch needs user growth through quality content and long-tail engagement.

Drawbacks?

This design algorithm isn’t without limitations. Some users may claim the implementation gives unfair advantage to specific viewers, allowing the Streamer to more easily recognize their comments. Other arguable points:

• Gives undeserved attention to users, not fair to all commenters as to which comments are promoted to the Streamer
• Does not follow conversations, only follows individual comments
• Possibility of certain viewer types (paying viewers) to achieve a higher level of chat engagement, due to their comments being promoted more to the Streamer. AKA: Pay-to-Win
• Lower commenting rate due to viewers awareness of the algorithm, which doesn’t promote their comments if they comment too frequently

Ensuring comment promotion isn’t too heavy-handed should accommodate for these. We want to make a richer chat experience for both sides so testing early and iterating on the algorithm will be important. Giving Streamers a setting for turning this feature on or off could provide a good middle ground upon release. If necessary, even more granular settings control could be given, allowing Streamers to decide which comments they preferred to see first and highlighted in their comment viewing window.

When posed the question “How do we grow as a community?” or “What ways can we build a more engaged user base?” it’s helpful to start with very simple user experiences. Chat for Twitch is an integral part of the whole streaming experience, yet small changes can drastically influence how both content producers and their audience feels during moments that matter most.

What are your thoughts? Would this algorithm and subsequent design help Twitch chat?