Abstract: Currently, most of the gas accident emergency response plans are stored in the form of text, and the on-site disposal personnel need to check a large number of texts to determine the corresponding disposal measures, which can hardly meet the rapidity and timeliness of emergency response. In order to overcome the shortcomings of poor timeliness and improve the reusability and flexibility of the gas emergency response plan stored in the form of text, an auxiliary decision-making framework based on the knowledge graph of gas accident emergency response was built, using a "top-down" approach. BERT-BiLSTM-CRF method is used to extract the background information of gas accident cases, and a method combining semantic role labeling with dependency parsing is proposed to build entity-relationship triad. The extracted gas emergency disposal entities and relational knowledge are finally stored and displayed in the Neo4j graph database. This study can provide effective support for auxiliary decision-making of gas accident emergency disposal.

Key words: gas emergency disposal, knowledge graph, BERT-BiLSTM-CRF model, dependency parsing, semantic role labeling