Document-level event extraction framework based on prompt learning

Currently, most research on event extraction adopts a fine-tuning approach to pre-trained language models based on downstream tasks, which may lead to inconsistencies between upstream and downstream tasks of the pre-trained models. To address these issues, a document-level event extraction model based on prompt learning is proposed. First, a named entity recognition module based on a multi-head attention mechanism is used to identify entities. Then, to avoid the use of trigger words and reduce data annotation difficulty, a strategy is proposed to quantify the importance of different argument roles for event types. Finally, the event argument extraction task is restructured into a conditional text generation task through a prompt-based event argument extraction module. Experimental comparisons are conducted on three publicly available document-level event extraction datasets, demonstrating the effectiveness of the proposed framework.