面向危险化学品事故的事理图谱构建

doi:10.16265/j.cnki.issn1003-3033.2024.05.1570

摘要/Abstract

摘要：

为提高危险化学品企业安全管理水平,构建危险化学品事故因果关系事理图谱。首先,依据事故调查报告,提出一种基于改进CasRel的实体关系联合抽取模型,该模型通过关系感知双向编码器表征法(R-Bert)和Span指针网络提高文本信息的抽取准确性;其次,利用相似度计算方法进行事件泛化,使得图谱更加完整和准确,并将其存储在图数据库Neo4j中,以可视化的方式直观展示事件之间的关联;最后,基于构建的事理图谱,设计出与之匹配的问答系统架构,开发危险化学品事故因果关系智能问答系统。研究结果表明:改进的CasRel模型计算结果F₁值为90.5%,较未改进模型准确率提升约2%。构建的危险化学品事故事理图谱及智能问答系统,在多个评价指标表现良好,能够清晰揭示事件之间的逻辑关系;所提方法可满足危险化学品事故知识问答需求,协助探究事故发生的规律和潜在风险因素,推断事故发展趋势。

关键词: 危险化学品事故, 事理图谱, 知识抽取, 因果关系, 智能问答系统

Abstract:

The causality causal graph of hazardous chemical accidents was developed to improve the safety management level of hazardous chemical enterprises. Firstly, based on the accident investigation report, an entity-relationship joint extraction model was proposed through an improved CasRel technique. Furthermore, the proposed model aimed to improve the extraction accuracy of textual information by incorporating the relationship-aware bidirectional encoder representation method (R-Bert) and Span pointer network. Subsequently, similarity calculation methods were used to generalize the events to enhance the graph's comprehensiveness and accuracy. Then, the refined data was stored in the Neo4j graph database visualizing the associations between events. Finally, the corresponding guestion-answering system was proposed based on the developed causal graph, and then an intelligent question-answering system for the causality of hazardous chemical accidents was proposed. The results indicated that the F₁ value calculated by the improved CasRel model was 90.5%, and the prediction accuracy of the proposed model was 2% higher than that simulated by the original model. The hazardous chemical accidents causal graph and intelligent question-answering system performed well in terms of multiple evaluation indexes, clearly revealing the logical relationship between events. Therefore, the proposed model in this study can meet question-answering needs of hazardous chemical accidents, facilitating the exploration of accident patterns and potential risk factors, and enabling accident trend prediction.

Key words: hazardous chemical accidents, casual graph, knowledge extraction, causality, intelligent question-answering system

中图分类号:

X937

李红儒, 栾婷婷, 邓明月, 陈文涛, 张雪. 面向危险化学品事故的事理图谱构建[J]. 中国安全科学学报, 2024, 34(5): 195-203.

LI Hongru, LUAN Tingting, DENG Mingyue, CHEN Wentao, ZHANG Xue. Development of causal graph for hazardous chemical accidents[J]. China Safety Science Journal, 2024, 34(5): 195-203.

图/表 15

图1

图2

图3

表1

图4

图5

图6

表2

图7

图8

图9

图10

图11

图12

表3

参考文献 20

[1]	刘成军. 炼油化工装置水锤事故分析及处理措施[J]. 石油与天然气化工, 2018, 47(3):101-107.
	LIU Chengjun. Analysis and treatment measures of water hammer accidents occurring in refining & chemica plants[J]. Chemical Engineering of Oil & Gas, 2018, 47(3): 101-107.
[2]	SINGHAL A. Introducing the knowledge graph: things, not strings[EB/OL]. (2012-05-16). https://www.blog.google/products/search/introducing-knowledge-graph-things-not/.
[3]	ZHANG Qi, WEN Yuanqiao, ZHOU Chunhui, et al. Construction of knowledge graphs for maritime dangerous goods[J]. Sustainability, 2019, 11(10):DOI:10.3390/su11102849.
[4]	龙丹冰, 魏君豪, 杨成. 知识图谱改进的施工行为安全风险与危险位置识别算法[J]. 中国安全科学学报, 2021, 31(12):10-16. doi: 10.16265/j.cnki.issn 1003-3033.2021.12.002
	LONG Danbing, WEI Junhao, YANG Cheng. An improved algorithm for construction behavior safety risk and hazard location identification based on knowledge graph[J]. China Safety Science Journal, 2021, 31(12): 10-16. doi: 10.16265/j.cnki.issn 1003-3033.2021.12.002
[5]	刘焕勇, 薛云志. 我们的实践:事理图谱,下一代知识图谱[EB/OL].(2018-12-25). https://blog.csdn.net/lhy2014/article/details/85247268.
[6]	YANG Congcong, SHI Xiaodong, WEI Chihping. Discovering event evolution graphs from news corpora[J]. IEEE Transactions on Systems, Man, and Cybernetics-Part A: Systems and Humans, 2009, 39(4): 850-863.
[7]	WANG Hong, ZHU Han, BAI Yunqing, et al. Construction of causality event evolutionary graph of aviation accident[C]. The 5^th International Conference on Transportation Information and Safety, 2019: 692-697.
[8]	王洁宁, 朱妍. 基于事理图谱的飞行冲突事故演化研究[J]. 安全与环境学报, 2023, 23(6):1961-1969.
	WANG Jiening, ZHU Yan. Research on the evolution of flight conflict accidents based on event logic graph[J]. Journal of Safety and Environment, 2023, 23(6): 1961-1969.
[9]	张鹏翔. 多维字符特征表示的铁路设备事故信息抽取方法[J]. 中国安全科学学报, 2022, 32(6):109-114. doi: 10.16265/j.cnki.issn1003-3033.2022.06.2732
	ZHANG Pengxiang. Information extraction method for railway equipment accidents based on multidimensional character feature representation[J]. China Safety Science Journal, 2022, 32(6): 109-114. doi: 10.16265/j.cnki.issn1003-3033.2022.06.2732
[10]	GRUBER T R. Toward principles for the design of ontologies used for knowledge sharing?[J]. International Journal of Human-computer Studies, 1995, 43(5/6): 907-928.
[11]	李跃艳, 王昊, 邓三鸿, 等. 面向事件本体的医学文本语义关联化研究[J]. 情报学报, 2022, 41(5):497-511.
	LI Yueyan, WANG Hao, DENG Sanhong, et al. Research on semantic relevance of medical text oriented to event ontology[J]. Journal of the China Society for Scientific and Technical Information, 2022, 41(5):497-511.
[12]	LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[C]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 42(2): 318-327.
[13]	拓雨欣, 薛涛. 融合指针网络与关系嵌入的三元组联合抽取模型[J]. 计算机应用, 2023, 43(7):2116-2124. doi: 10.11772/j.issn.1001-9081.2022060846
	TUO Yuxin, XUE Tao. Joint triple extraction model combining pointer network and relational embedding[J]. Journal of Computer Applications, 2023, 43(7):2116-2124. doi: 10.11772/j.issn.1001-9081.2022060846
[14]	海南省应急管理厅. 三门峡市河南煤气(集团)有限责任公司义马气化厂“7·19”爆炸事故调查报告[EB/OL]. (2020-08-10). http://yjglt.hainan.gov.cn/xxgk/sgdcbg/202008/t20200810_2831571.html.
[15]	葛任贤. 基于互联网知识抽取的汽车故障事理图谱构建方法[D]. 广州: 广东技术师范大学, 2023.
	GE Renxian. The construction method of automobile fault event-logic graph based on internet knowledge extraction[D]. Guangzhou: Guangdong Polytechnic Normal University, 2023.
[16]	张维宁, 申喜凤, 李美婷, 等. 融合关系标签和位置信息的中文医疗文本因果关系抽取方法研究[J]. 医学信息学杂志, 2024, 45(1):21-26.
	ZHANG Weining, SHEN Xifeng, LI Meiting, et al. Study on the method of causality extraction from Chinese medical texts by integrating relational label and location information[J] Journal of Medical Informatics, 2024, 45(1):21-26.
[17]	ZHENG Suncong, WANG Feng, BAO Hongyun, et al. Joint extraction of entities and relations based on a novel tagging scheme[C]. the 55^th Annual Meeting of the Association for Computational Linguistics, 2017: 1227-1236.
[18]	ZENG Xiangrong, ZENG Daojian, HE Shizhu, et al. Extracting relational facts by an end-to-end neural model with copy mechanism[C]. the 56^th Annual Meeting of the Association for Computational Linguistics, 2018:506-504.
[19]	FU T J, LI P H, MA Weiyun. GraphRel: modeling text as relational graphs for joint entity and relation extraction[C]. the 57^th Meeting of the Association for Computational Linguistics, 2019: 1409-1418.
[20]	WEI Zhepei, SU Jianlin, WANG Yue, et al. A novel cascade binary tagging framework for relational triple extraction[C]. the 58^th Annual Meeting of the Association for Computational Linguistics, 2020: 1476-1488.

数据集	训练集	验证集	测试集
数据集/条	10 280	2 560	2 560
三元组数量/条	56 930	10 990	10 690

模型	P/%	R/%	F₁/%
去掉R-bert	89.1	88.1	88.3
去掉Span	90.1	88.7	87.2
改进的CasRel模型	90.6	91.8	90.5