Modeling Story Expectations: A Generative Framework using LLMs

arXiv cs.CL · 18 hours ago

Researchers developed a framework using large language models to model consumer expectations about narrative stories by generating multiple possible continuations and extracting features like emotion and narrative paths. The method was validated through survey data comparing LLM predictions to human-reported beliefs and through observational data from an online reading platform, with validation across multiple narrative features. The approach enables scalable prediction of reader engagement based on forward-looking expectations about story outcomes, with applications for content creation and platform strategy.