Abstract: The study of event coreference is a fundamental task in the field of natural language processing. To investigate coreference between all events in open domains, an approach that defines event coreference directly from the event expressions is proposed, with a specific focus on verb-driven events. A guideline for verb-driven event coreference is further composed, which divides event coreference into strong and weak coreference. Then, a manually annotated Chinese corpus consisting of over 380 thousand characters is constructed, comprising 1,000 documents, 29,250 verbs, 1,858 strong coreference chains and 994 weak coreference verb pairs. A pipeline experiments for coreference resolution is conducted, which demonstrate an F₁ score of 96.9 in mention detection phase, and average F₁ scores of 50.0, 64.4, and 71.7 for coreference prediction via prompt engineering, instruction tuning and knowledge distillation of LLM, respectively.