基于指针标注的金融事件联合抽取模型  

A joint extraction model for financial events based on pointer annotation

作　　者：冯珊珊 赵辉[1] 曹亚亚 FENG Shanshan;ZHAO Hui;CAO Yaya(School of Computer Science&Engineering,Changchun University of Technology,Changchun 130102,China)

机构地区：[1]长春工业大学计算机科学与工程学院,吉林长春130102

出　　处：《长春工业大学学报》2023年第5期441-448,共8页Journal of Changchun University of Technology

基　　金：吉林省教育厅“十三五”科学技术项目(JJKH20200677KJ)。

摘　　要：针对金融领域事件抽取存在的元素重叠、语句歧义以及级联错误传播问题,提出一种基于指针标注的金融事件联合抽取模型,首先利用PERT预训练模型提取语义信息;再融入事件类型信息进行语义增强,SATT-BILSTM模型加强特征提取,指针标注解决元素重叠;最后为了增强模型的鲁棒性和泛化性,在模型训练的过程中加入了对抗训练。文中提出的模型在FewFC测试集上进行对比实验,结果表明,抽取效果更佳。To address the problems of element overlap,statement ambiguity and cascading error propagation in event extraction in the financial domain,a joint extraction model of financial events based on pointer annotation is proposed,which first extracts semantic information by using PERT pre-training model,then incorporates event type information for semantic enhancement,SATT-BILSTM model to enhance feature extraction,pointer annotation to solve element overlap,and finally,to enhance the model's finally,in order to enhance the robustness and generalization of the model,adversarial training is added in the process of model training.The model proposed in this paper is tested on the FewFC test set for comparison,and the experimental results show that the ex-traction effect is better.

关 键 词：联合学习 元素重叠 语句歧义 错误传播 指针标注 对抗训练 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]