Why Twitch Chat?

We’re interested in exploring the interactive chat culture of one of the most popular streaming platforms, Twitch. With the immense customizability of features and contents, as well as the means of engagement, Twitch chat allows the users to not only be the passive viewer of streams, but active participants with the streamers and the fellow members of the community as well. This very essential part of the Twitch experience is the core to the video games communities on the platform, however often times harmed by toxic chat behaviors.

With this visualization, we want to

Allow for deeper understandings of Twitch chat culture and conversations around different aspects of chat elements.

Find meaningful relationships between toxicity in chats and Twitch engagements.

Give the general public an understanding of livestreaming culture and the issues involved.

An Interactive Introduction

Let's take a closer look at what goes into a Twitch chat messsage and how we might extract a sentiment from it.

We sample some chats from a 2-hour long stream from Twitch user "Pokimane", presented below. Scroll through to start seeing the messages and how we process them. Don't scroll too fast, and try not to use the scroll wheel on a mouse! (scroll wheels can jump very far down the page at once and miss some of the annotations in between).

.
.
.
This is the first potentially negative message we've seen. Our automated sentiment analysis tool has judged the previous messages to be positive or neutral; this message is judged as potentially negative because of the innuendo/language used. A moderator might have this message flagged as necessary to review.
.
.
This is an example of a message that requires special processing to understand. "Boosted" is slang with specific meaning on Twitch, implying that somebody faked their way to a higher ranking. We had to compile slang terms like "boosted" and tell our sentiment analyzer to recognize such terms.
.
.
Common emotes on twitch such as PepeLaugh , LUL , KEKW , etc. are used to denote laughter - but laughter can be sarcastic and mocking or genuine and positive. One weakness of sentiment analysis is that without further context, some messages cannot be clearly classified as positive or negative. We err on the side of neutrality for this analysis.
.
This is an example of a message that is interpreted as extremely positive. Though we have cut it off for the sake of brevity, it's the same message spammed dozens of times. Such "spam" is common on Twitch, both positive and negative.
.
.
.
.
.
.
.
Bullying is quite common Twitch; there will inevitably be people who enter streams just to say mean things to the streamer (who is often reading these messages as they play). The sentiment analysis tool attempts to find such insults, which it knows to mark as negative.
Another instance of "positive" spam. Here, they are using the pokiWK emote, meant to represent a WK ("white knight") cheering her on, as well as the slang term "hyperclap". We have manually compiled a list of positive emotes and slang for our sentiment analyzer to use.
.
.
.
.
...almost at the end now...
.
.
.
PokiHYPERS ! At the end of this series of messages, we can see that there are approximately the same number of positive and negative messages. However, the cumulative sentiment score is overwhelmingly positive. This reflects the general attitude of users in Pokimane's chat - they are likely to be very positive through sharing supportive emotes, cheering her on, and even spamming messages of positivity.



We now move to a new streamer: Tyler1. His chat is far different from Poki's.
.
Tyler1 is known for being more of a "toxic" streamer. He attracts a far larger male audience, cultivates a more angry persona, and enjoys yelling and screaming as he plays. Let's take a look at how his chat evolves.
.
.
This message casts a negative light on the chat, calling them "low elo", which is slang for unskilled. Such self deprecation and insults are common in his chat.
.
.
As you can see by now, there is more reference to sexual and immature content. Our sentiment analysis tool tries to flag this sort of content negatively, since it is important to moderate these types of messages.
.
Frequent and nonsensical spam is also common in Tyler1's chats. While not necessarily offensive, it is deliberately meant to annoy people and cause a reaction. We attempt to tag such spam as negative with our sentiment analyzer.
.
Profanity is also common. Needless to say, this is considered strongly negative.
.
.
.
.
.
.
.
Many such messages have no real words, but instead use slang and serve as pure reaction. "Monka" typically refers to expressions featuring Pepe the Frog; this is usually verbal slang, but users with a special extension see emotes like or , denoting fear/stress/sadness.
.
.
.
Despite not making an insulting statement, messages like this can still create a culture of negativity that persists in this streamer's chatroom.
.
.
.
.
... almost at the end...
.
.
.
.
More negative spam...
.
Good question, anonymous user. The answer is: you don't! (and yes, this is a real chat log from Tyler1's stream; it seems like even viewers are bewildered sometimes).



As you could see from the past two chat logs, people generally don't interact or respond to each other on Twitch chats. Each chat message individually seems random and chaotic; in aggregate, however, we can get a sense for the overall culture and attitude of each streamer's community.
See the difference? Tyler1 has a higher quantity of negative messages than Pokimane . Moreover, Poki has more positive sentiment than the proportion of messages would have you believe, showing that each of Poki's positive messages are very intensely positive - likewise, Tyler1's negative messages are very intensely negative.



Scroll down more to see more detailed charts for each streamer.

Let's take a closer look into each streamer's chat sentiments

Pokimane, HealthyGamer_GG & loltyler1 all have such differently perceived personas, content, and styles, but how are those differences reflected in sentiments? What are some of the key events and elements that drive changes and collective actions in Twitch chat? How about some interesting breakdowns that we’re able to acquire based on sentiment analysis? In this following sections, you'll be able to explore the different aspects of 2-hour randomly-chosen streaming content from each streamer, especially those that are unware of from a viewer's perspective.

Tips

Use the tabs to navigate between different streamers' chat breakdown.

#1 HealthyGamer_GG #2 Pokimane #3 loltyler1
#1 HealthyGamer_GG #2 Pokimane #3 loltyler1
  • Streamer
    • Pokimane
  • Primary Category
  • Variety
  • Log Date
  • April 18th, 2021
    • Link to stream
  • Theme in Abstract
  • OfflineTV Valorant Tournament
#1 HealthyGamer_GG #2 Pokimane #3 loltyler1
  • Streamer
    • loltyler1
  • Primary Category
  • League of Legends
  • Log Date
  • May 6th, 2021
    • Link to stream
  • Theme in Abstract
  • nice weather League play
Pokimane loltyler1 HealthyGamer_GG
Back To Top