Hybrid Annotation for Propaganda Detection: Integrating LLM Pre-Annotations with Human Intelligence