Event Extraction Model Based on Dependency Syntax and Label Pointer Network

Event extraction is an important and challenging task in information extraction, and a key step in the construction of a eventic graph, which is designed to extract specific event information from text. News events can usually be classified as normal, overlapping and nesting events according to how they occur. At this stage, majority of the research focuses on the extraction of normal events and ignores the extraction of overlapping and nesting events. Additionally, the textual syntax of events is structurally complex and suffers from trigger word ambiguity, that is, trigger words may carry multiple meanings, making it difficult to determine whether they are the trigger word for an event or something else in a given context. For addressing the above problems, this paper designs a joint event extraction model DS-LPN, which consists of three modules, namely, text encoding layer, dependency feature extraction layer, and label pointer network layer. Experiments show that the model’s performance on the FewFC dataset in the Chinese financial domain is significantly improved compared to the baseline model.