Abstract
News articles often reference numerous organizations, but traditional NamedEntity Recognition (NER) treats all mentions equally, obscuring which entitiesgenuinely drive the narrative. This limits downstream tasks that rely onunderstanding event salience, influence, or narrative focus. We introduceProtagonist Entity Recognition (PER), a task that identifies the organizationsthat anchor a news story and shape its main developments. To validate PER, wecompare he predictions of Large Language Models (LLMs) against annotations fromfour expert annotators over a gold corpus, establishing both inter-annotatorconsistency and human-LLM agreement. Leveraging these findings, we usestate-of-the-art LLMs to automatically label large-scale news collectionsthrough NER-guided prompting, generating scalable, high-quality supervision. Wethen evaluate whether other LLMs, given reduced context and without explicitcandidate guidance, can still infer the correct protagonists. Our resultsdemonstrate that PER is a feasible and meaningful extension tonarrative-centered information extraction, and that guided LLMs can approximatehuman judgments of narrative importance at scale.