建设工程事故文本的知识发现:以PPE类不安全行为为例

Knowledge Discovery of Construction Accident Texts:An Example of Unsafe Behavior Related to PPE

为了丰富建设工程领域的安全知识,从事故文本中挖掘和发现施工人员的不安全行为,以个人防护用品PPE类不安全行为为例,采用基于规则的自然语言处理方法,从事故文本中自动抽取此类不安全行为。从政府官网等收集195份建设工程事故调查报告作为文本挖掘语料,通过哈尔滨工业大学的语言技术平台LTP展开词法分析和依存句法分析,构建PPE类不安全行为的11条抽取规则并确定抽取流程。再以网络爬虫收集的427份事故调查报告展开实例应用,按照流程自动抽取PPE类不安全行为。结果表明:平均抽取准确率为94.70%,召回率为67.57%。研究能够为建设工程事故文本的知识发现提供理论启示和实践路径。

To enrich the safety knowledge in the field of construction industry,the unsafe behaviors of workers are mined anddiscovered from accident texts.This paper took the unsafe behavior related to personal protective equipment(PPE)as an example,and this kind of unsafe behavior was automatically extracted from accident texts using a rule-based natural language processingmethod.195 construction accident investigation reports were collected from government websites to form a text mining corpus.Thelexical analysis and dependency parsing of the texts were carried out through the Language Technology Platform(LTP)of HarbinInstitute of Technology.The 11 extraction rules of the unsafe behavior related to PPE were constructed and the extraction process isdetermined.Then,another 427 construction accident investigation reports collected by web crawlers were used as examples toautomatically extract the unsafe behavior related to PPE according to the extraction process.The results show that the averageextraction accuracy is 94.70%and recall rate is 67.57%.The study can provide a theoretical inspiration and practical path forknowledge discovery in construction accident texts.