基于双向长短时记忆网络的地铁应急知识抽取与推理

Metro Emergency Knowledge Extraction and Knowledge Reasoning Based on BiLSTM-CRF

摘要:

为解决文本类地铁应急处置流程存在的流程顺序关系不明确、流程执行人员模糊等问题，提出了基于BiLSTM-CRF（Bidirectional Long Short-Term Memory-Conditional Random Field）的地铁应急处置知识抽取与推理方法。首先，利用BiLSTM-CRF方法对地铁应急处置流程的文本资料进行命名实体识别，完成文本资料的知识抽取；其次，选用TransD模型对识别后实体数据进行知识推理，从而完成以实体和属性对为节点、关系对为边的知识图谱构建；最后，利用Neo4j图数据库对构建的地铁应急处置流程知识图谱进行了可视化展示和案例分析。研究结果表明，基于BiLSTM-CRF的知识抽取模型的精确率、召回率和F1值均达到了90%以上，且基于BiLSTM-CRF的TransD模型的推理结果准确率提升了22.92%，保证了知识图谱构建的准确性，可为地铁应急管理提供决策支持。

Abstract:

To address issues such as the unclear sequence of procedures and ambiguity in the personnel responsible for executing the emergency response procedures in text-based metro emergency response processes， this paper proposes a knowledge extraction and knowledge reasoning method for metro emergency response procedures based on knowledge graph of bidirectional long short-term memory- conditional random field （BiLSTM-CRF）. First， the BiLSTM-CRF method is used to identify the named entity of the text data of the metro emergency response process， and complete the knowledge extraction of the text data. Then， the TransD model is selected to conduct knowledge inference on the identified entity data， thereby completing the construction of a knowledge graph with entities and attribute pairs as nodes and relational pairs as edges. Finally， the Neo4j graph database is used to visualize and analyze the knowledge graph of metro emergency response process. The research results show that the precision， recall， and F1 value of the knowledge extraction model based on BiLSTM-CRF have all reached more than 90%， and the accuracy of the inference results of the TransD model based on BiLSTM-CRF has increased by 22.92%， ensuring the accuracy of knowledge graph construction and providing decision support for subway emergency management.