基于深度学习的演训文书事件提取技术  

Event extraction technology of training documents based on deep learning

作　　者：贾春秋 窦林涛[1] 曾清[1] JIA Chunqiu;DOU Lintao;ZENG Qing(Jiangsu Automation Research Institute,Lianyungang 222061,China)

机构地区：[1]江苏自动化研究所,江苏连云港222061

出　　处：《指挥控制与仿真》2025年第2期68-74,共7页Command Control & Simulation

摘　　要：演习训练过程中会产生海量的文本信息,其复杂性和多样性给训练评估人员带来巨大的认知压力,如何充分挖掘演训文书中的非结构化数据,为分析评估人员提供高效的服务,是演训数据处理的重难点问题。针对演训文书中专业名词多、中英文共存、短句关键信息密集等特点,提出一种基于深度学习的事件提取技术。结合ALBERT强大的文本特征提取、CRF序列标注的结构化预测等优势,构建了演习训练文书方面的事件提取模型。为检验模型的性能,在演训数据集上进行实验,实验结果表明,利用该模型文本提取效果良好,对从演训文书提取信息来说,具有一定的应用意义。A large amount of textual information will be generated in the exercise training process,and enormous cognitive pressure will be exerted on training evaluators due to complexity and diversity of such information.How to fully extract unstructured data from exercise training documents and provide efficient services for analysis and evaluation personnel is a challenging issue in data processing.In this paper,we propose a deep learning-based event extraction technique for exercise training documents,which addresses the characteristics of abundant professional terminology,coexistence of Chinese and English,and dense key information in short sentences.By leveraging the powerful text feature extraction of ALBERT and the structured prediction of CRF sequence labeling,we construct an event extraction model for exercise training documents.Experimental results on the training data set demonstrate that this model performs well in text extraction and has practical applications for extracting information from exercise training documents.

关 键 词：深度学习 演训文书 事件提取 

分 类 号：E919[军事]