通过聚合事件元素信息的篇章级事件抽取

he current financial field has a large amount of text data, and quickly extracting event related information from the text can help people quickly grasp the overview of financial activities. Currently, more financial text data is presented in the form of long text, and traditional sentence level event extraction methods perform poorly on discourse level financial texts. Based on this situation, there have been many research works on text level event extraction applied to financial texts in recent years. The purpose of text level event extraction is to extract structured and complete text level event information from documents, mainly providing the ability to solve the problem of scattered event elements and multiple events that sentence level event extraction methods lack. This article proposes a discourse level event extraction model that aggregates event element information to address the aforementioned issues. This method can fuse and encode the information of long texts for event element extraction, and construct inter entity dependency information for discourse level text encoding. The experimental results on the publicdataset ChFinAnn show that the F1 value of the model proposed in this paper reaches 78.3%.