山西大学学报（自然科学版）2025，Vol.48Issue(5)：871-879,9.DOI:10.13451/j.sxu.ns.2025013
面向危险化学品事故的事理图谱抽取方法研究
Research on the Construction Method of Eventic Graph of Hazardous Chemicals Accidents
摘要
Abstract
In order to prevent the occurrence of chemical accidents derived from storage and transportation accidents of hazardous chemicals,the accident risk identification process is optimized to reduce the reliance on manual work.Firstly,from the perspective of Chinese corpus,based on the incorporation of many semantic information,the hazardous chemicals accident entities are extracted by Multi-Granularity Dilated Convolution Networks using Co-Predictor Layer to infer the relationship between words;then,for the entities that have been extracted,positional coding is incorporated to recognize the object of each subject as well as the correspond-ing multiple relations to achieve the Hazardous Chemicals Incident Relationship Recognition.The results show that the F1 value of the Namedity Extraction model reaches 92.91%and 96.59%in the self-constructed and public datasets,respectively,and the F1 val-ue of the Relation Recognition model reaches 76.04%;the performance of the two models improves compared with the existing methods,and the Relation Recognition model,in particular,has a clear advantage and achieves a performance lead of 6.28%.Named Entity Recognition of hazardous chemicals and Relation Extraction further establishes the hazardous chemicals Eventic Graph and early warning system.关键词
多粒度空洞卷积网络/联合预测层/实体抽取/位置编码/关系识别Key words
multi-granularity dilated convolution networks/co-predictor layer/named entity recognition/positional encoding/rela-tion extraction分类
信息技术与安全科学引用本文复制引用
陈观林,孔振华,汪俊霖,翁文勇..面向危险化学品事故的事理图谱抽取方法研究[J].山西大学学报（自然科学版）,2025,48(5):871-879,9.基金项目
浙江省重点研发计划项目(2020C03091) （2020C03091）
