关键词: 事件检测, 句法信息, 无参数, 无触发词, 类型感知器

Abstract: Abstract: Event Detection (ED) is one of the most important tasks in the field of information extraction, aiming to identify instances of specific event types in text. Existing event detection methods usually use adjacency matrix to express syntactic dependency, but the adjacency matrix often needs to use Graph Convolutional Network (GCN) to encode to obtain syntactic information, thereby increasing the complexity of the model. Therefore, we propose a event detection model that convert the parent word and its context into a position marker vector, and at the source of the model, the parent word information is incorporated into the word embedding of the child word in a parameter-free manner to strengthen the semantic representation of the context, without to go through the GCN network for encoding . In addition, considering that the tagging of trigger words is time-consuming and laborious, this paper designs a type perceptron based on a multi-head attention mechanism, which can model potential trigger words in sentences to achieve event detection without triggers. In order to verify the performance of the proposed method, this paper conducted experiments on the ACE2005 data set and the low-resource Vietnamese data set. Among them, compared with the event detection model GTN-ED on the ACE2005 data set, the F1 value increased by 3.7%, and compared with the two-class baseline model TBNNAM, the F1 value has increased by 9%. The results show that the integration of syntactic information into Transformer can effectively connect the scattered event information in the sentence to improve the accuracy of event detection.

Key words: event detection, syntactic information, without parameters, without triggers, type perceptron