An Analysis of Interaction and Engagement in YouTube Live Streaming Chat

Live streaming is a popular social media platform. Most live streaming services allow viewers to interact with each other and the broadcaster via text chat. Thus, exploring user behavior and chat communication within a live streaming context is increasingly important. In this paper, we explore YouTube general live streaming chat. We extracted a corpus in the political domain from Trump's 2020 presidential campaign in the United States. Then, using advanced Natural Learning Processing (NLP) algorithms, we examined users' behavior in live chats. We focused on three elements of YouTube user behaviour: chat content, commentators' behaviour, and user engagement evoking factors. We observed that the chat messages are very emotional. They include a lot of emojis, some of which are domain-depended. Almost all messages express sentiment and the positive sentiment outweighs the negative. Abusive language is very common in the messages. However, while heavy users express more sentiment, the tend to use less abusive language. Although it is difficult to know exactly what was said in the video that caused the engagement of the chat participants, using a topic model, we were able to show that sarcasm evokes involvement.