基于span指针网络的危险化学品安全知识抽取  

Safety Knowledge Extraction of Hazardous Chemicals Based on Span Pointer Network

作　　者：许蒙蒙 毛帅[1] 唐漾[1] 王冰[1] XU Meng-meng;MAO Shuai;TANG Yang;WANG Bing(Key Laboratory of Smart Manufacturing in Energy Chemical Process,Ministry of Education,East China University of Science and Technology,Shanghai 200237,China)

机构地区：[1]华东理工大学能源化工过程智能制造教育部重点实验室,上海200237

出　　处：《控制工程》2022年第6期1082-1089,共8页Control Engineering of China

基　　金：国家科技部重点研发计划项目(2018YFC0809302)。

摘　　要：危险化学品安全知识存在数据长、实体跨度大、语义复杂度高且部分实体间存在嵌套等特点,这些特点增大了危险化学品安全知识的抽取难度。为了有效地解决危险化学品安全知识的抽取问题,并针对关系抽取任务中流水线方法存在的误差累积和实体冗余问题,提出了一种基于span指针网络的关系联合抽取方法,旨在抽取危险化学品安全知识中的三元组信息。此外,为了提升模型的抽取性能和泛化能力,在span指针网络的模型基础上,添加了对抗训练和Lookahead优化器。最终的实验结果证明,此方法在解决危险化学品安全知识的抽取问题上具有优越的性能。Safety knowledge of hazardous chemicals has the characteristics of long data, large entity span, high semantic complexity, and nesting among some entities. These characteristics increase the difficulty of extracting safety knowledge of hazardous chemicals. In order to effectively solve the problem of extracting safety knowledge of hazardous chemicals, and the problems of error accumulation and entity redundancy of the pipeline method in the relation extraction task, a method of relation joint extraction based on span pointer network is proposed in this paper, which aims to extract the triple information in safety knowledge of hazardous chemicals. In addition, in order to improve the extraction performance and generalization ability of the model, adversarial training and Lookahead optimizer are added to the span pointer network model. The final experimental results prove that this method has superior performance in solving the problem of extracting safety knowledge of hazardous chemicals.

关 键 词：span指针网络 危险化学品安全知识 关系抽取 Lookahead优化器 对抗训练 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]