📑 arXiv 2d ago
DiZiNER: Disagreement-guided Instruction Refinement via Pilot Annotation Simulation for Zero-shot Named Entity Recognition
DiZiNER simulates pilot annotation processes where multiple heterogeneous LLMs act as annotators and supervisors to refine instructions for zero-shot NER. The framework identifies systematic errors by generating disagreements between models, mirroring how human annotation resolves inconsistencies to improve zero-shot performance toward supervised baselines.