基于提示学习的文档级事件抽取方法

he main task of event extraction is to automatically extract entities, event types and their relationships related to specific events from unstructured text. At present, the research methods of event extraction mostly adopt the method of fine-tuning according to the downstream tasks on the pre-training language model, which will lead to inconsistency between the upstream and downstream tasks of the pre-training model; Moreover, most of the researches still focus on the event extraction within the scope of a single sentence. Such extraction results have the problems of incomplete event information and argument omission. Aiming at the above problems, a document level event extraction model based on prompt learning is proposed. First, the named entity recognition module based on multi-head attention mechanism is used to identify the entities in the document. Then, in order to avoid the use of trigger words and reduce the difficulty of data annotation, a strategy to quantify the importance of different argument roles to event types is proposed, which is used to locate the event center sentence in the document and determine the event type. Finally, through the event argument extraction module based on prompt learning, the argument extraction task is reconstructed into a conditional text generation task to achieve argument extraction. Finally, experiments and comparisons are carried out on three open document-level event extraction data sets, which prove the innovation and effectiveness of this method.