Developing Feedback Systems For Live Stream Communities Using Large Language Models

Abstract

Live streaming has become a key mode of community engagement, enabling real-time interaction between creators and audience. Platforms like Twitch have transformed content creation into a participatory, performative practice that blends entertainment, emotional labor, and community building. Despite its cultural and economic significance, little research has examined how streamers interpret and act on audience feedback in ways that support sustained growth, well-being, and meaningful interaction. This dissertation explores how large language models (LLMs) can power intelligent feedback systems that help streamers manage content production and community engagement. Through three studies, it examines feedback from multiple angles: streamers' existing information practices, proactive strategies for soliciting input, and reactive techniques for interpreting audience response post-stream. This research demonstrates how LLMs can synthesize audience input into insights that are actionable, emotionally supportive, and strategically meaningful. The work contributes both deployable systems and design principles for building feedback tools attuned to the demands of live stream communities.