基于命名实体识别的铁路应急预案智能管理方法

Intelligent Management Method for Railway Emergency Plan Based on Named Entity Recognition

铁路应急预案是应对自然灾害、事故灾难、社会安全等突发事件,快速做出科学决策、组织救援的行动指南。由于应急预案多以非结构化文档存储,针对其结构化及检索过程中信息损失的问题,提出基于命名实体识别的铁路应急预案智能管理方法。首先,应用排版样式、编辑语义、文本语义等信息,实现流式文档结构的智能解析、附件识别及时序文档的信息综合;然后,以运维知识图谱中实体及其类别构建实体类别集,嵌入实体类别信息并与字符嵌入拼接生成增强特征,对文本进行实体的命名标识及段属性标识;最后,采用递归神经网络并引入注意力机制构建命名实体层次表达,融合结构和语义因素实现检索结果的智能排序。实验证明,该方法能准确解析文档结构并在低标注下识别命名实体,实现应急预案的智能检索与排序,可有效提高铁路应急预案管理的智能化水平。

A railway emergency plan serves as an action guide for making fast efforts in scientific decision making and rescue organizing in response to emergencies such as natural disasters,accident disasters and social security events.As emergency plans are usually stored in non-structured documents,which are vulnerable to information loss in its structuring and retrieval process,an intelligent management method for railway emergency plans based on named entity recognition was proposed to solve the problem.Firstly,the intelligent analysis of the flow document structure,recognition of attachments,and information integration of the time-sequence documents were realized by applying information such as formatting style,editing semantics and text semantics.Then,an entity category set was constructed with entities and their categories in the knowledge graph,with the entity category information spliced with the character embedding to construct the enhanced feature vectors,and the entities extracted from texts were named and given paragraph attributes.Finally,a hierarchical expression of named entities was constructed by using a recursive neural network and an attention mechanism,with the document structure and semantics factors integrated to realize the retrieval and results ranking.The experimental results show that this method can accurately analyze the document structures and recognize named entities under low annotation conditions,which makes the emergency plans retrieved and sorted intelligently,thus improving the intelligent level of railway emergency plan management.