Method and system for extracting chapter-level event argument of document and medium

The invention discloses a text-level event argument extraction method and system of a document and a medium, belongs to the field of text information extraction, and aims at three retrieval modes of context consistency retrieval, mode consistency retrieval and self-adaptive mixed retrieval, a retrieval enhancement generation model based on a T5 model is constructed, a corresponding retrieval enhancement strategy is executed, and a text-level event argument is extracted. A sample that helps the demonstration model to solve the task can be recalled. According to the method, a reference vector is generated as a depth clue through an adaptive hybrid retrieval enhancement normal form, so that the analogy capability of the model can be improved. 本发明公开了一种文档的篇章级事件论元抽取方法、系统及介质，属于文本信息抽取领域，针对上下文一致性检索、模式一致性检索和自适应混合检索这三种检索方式，通过构建基于T5模型的检索增强生成模型，执行对应的检索增强策略，可以回忆起有助于演示模型应该如何解决任务的样例。其中，本发通过自适应混合检索增强范式来生成参考向量作为深度线索，能够提高模型的类比能力。