Document-level Overlapping Event Extraction via ChatGPT Leading Decision and Cross-Sentence Entity Representation Enhancements

Document-level event extraction is still a challenge when the document contains multiple events with multiple overlapped arguments. In the ChFinAnn dataset, over 37% of documents match this situation. However, existing works have not yet clearly interpreted the Overlapping Events. There is also no targeted solution to the problem. To address the above problems, in this paper, we first categorize and define the Overlapping Event, and then give a practical framework to solve the problem of Overlapping Event extraction. We propose a new metric called Argument Correlation to calculate the likelihood of overlapping event. We propose a parallel model DESIA-CHAT, which introduces ChatGPT as the leading model to distinguish Regular and Overlapping Events. Besides, for the issue of Cross-Sentence argument overlap, we enhanced the entity representation to capture complete features. We conducted experiments on two datasets(ChFinAnn and Assevent). Experimental results show that our method achieves new state-of-the-art performance on the 2 datasets.