交通事故致因知识图谱构建及风险因素挖掘  

Traffic Accident Causation Knowledge Graph Construction and Risk Factor Mining

作　　者：王占中[1] 张书源 杨萌 兰若冰 吴智豪 WANG Zhanzhong;ZHANG Shuyuan;YANG Meng;LAN Ruobing;WU Zhihao(Transportation College,Jilin University,Changchun 130025,China)

机构地区：[1]吉林大学交通学院,长春130025

出　　处：《同济大学学报(自然科学版)》2025年第4期611-618,共8页Journal of Tongji University:Natural Science

基　　金：吉林省自然科学基金面上项目(20230101112JC)。

摘　　要：利用交通事故调查报告中的数据,构建交通事故致因知识图谱并分析风险因素。首先,基于微调通用信息抽取统一框架预训练模型,构建适用于低数据量的交通事故致因命名实体识别模型,并生成实体集;其次,通过结构化处理和本体构建,利用图数据库Neo4j存储交通事故致因知识图谱,实现可视化;再次,基于专家经验和预训练语言文本分类模型,对交通事故致因实体进行标准化;最后,构建基于交通事故致因图谱的风险因素分析方法,通过分析标准化实体的类型分布和度分布,挖掘各因素对事故的触发特征与贡献,并进行关联规则挖掘。这些方法和分析结果提供了对历史事故风险因素的深入理解与探索。In this paper,we use data from traffic accident investigation reports to construct a traffic accident causation knowledge graph and analyze risk factors.Firstly,we construct the recognition model of named entities of traffic accident causation applicable to low data volume based on the fine-tuned UIE pre-training model for the generation of the entity set.Secondly,through the structured processing and ontology construction,the graph database Neo4j is used to store the traffic accident causation knowledge graph for visualization.Thirdly,based on the expert experience and pre-trained language text classification model,the traffic accident causation entities are standardized.Finally,a risk factor analysis method based on the traffic accident causation graph is constructed to mine triggering characteristics and contributions of each factor by analyzing the type distribution and degree distribution of standardized entities,and to perform the association rule mining.The results of these methods and analyses provide an in-depth understanding and exploration of historical accident risk factors.

关 键 词：交通运输 知识图谱 致因分析 数据挖掘 命名实体识别 

分 类 号：U491.31[交通运输工程—交通运输规划与管理]