Analyzing Audience Comments: Improving Interactive Narrative with ChatGPT.

This paper presents a novel method that utilizes ChatGPT for the categorization of audience comments in game live streams, treating it as a zero-shot task. Audience participation games have gained significant popularity in the realm of game live streaming, playing a vital role in game promotion and audience engagement. Streamers employ various techniques such as storytelling and interactive narrative to cultivate a larger fan base and enhance the value of their streams. Simultaneously, the audience generates diverse comments that directly impact the streamer’s interactive narrative and storytelling. However, the traditional methods for comment analysis in game live streams are lacking in terms of speed and cost-effectiveness. Therefore, our aim is to investigate whether ChatGPT can fulfill these requirements. Through experimental evaluation, our results indicate a majority choice of 54.34% and a human choice of 82.61%, showcasing that ChatGPT, when employed with suitable prompts, can address the aforementioned need.