Tag
This paper introduces MiGUE-Bench, a systematic benchmark for evaluating LLMs on multi-granularity event analysis, spanning single- to cross-document tasks including event detection, relation reasoning, structure induction, and future prediction.