Audience and Streamer Participation at Scale on Twitch

TL;DR

This study analyzes Twitch's audience and streamer interactions at scale, identifying five stream types and providing insights to improve engagement and platform design based on a large dataset.

Contribution

It introduces a comprehensive classification of Twitch streams based on size and interaction style, supported by analysis of over 12 million chat messages and 45 hours of video.

Findings

Identified five distinct stream types with unique interaction patterns

Provided data-driven design implications for improving engagement

Analyzed large-scale chat and video data to understand participation dynamics

Abstract

Large-scale streaming platforms such as Twitch are becoming increasingly popular, but detailed audience-streamer interaction dynamics remain unexplored at scale. In this paper, we perform a mixed-methods study on a dataset with over 12 million audience chat messages and 45 hours of streaming video to understand audience participation and streamer performance on Twitch. We uncover five types of streams based on size and audience participation styles: Clique Streams, small streams with close streamer-audience interactions; Rising Streamers, mid-range streams using custom technology and moderators to formalize their communities; Chatter-boxes, mid-range streams with established conversational dynamics; Spotlight Streamers, large streams that engage large numbers of viewers while still retaining a sense of community; and Professionals, massive streams with the stadium-style audiences. We discuss challenges and opportunities emerging for streamers and audiences from each style and conclude by providing data-backed design implications that empower streamers, audiences, live streaming platforms, and game designers