|
With the development of natural language processing technology, automatic text processing technology oriented to accident safety analysis is gaining more and more attention. By analyzing the passage of an accident text and reasoning out the causes and measures related to that accident, not only can the accident text be processed for automatic analysis, but also the analysis results can be used as safety knowledge to improve the safety awareness of staff. Therefore, we designed a text event reasoning model for safety analysis scenarios. The model is mainly divided into two parts: 1) a natural language inference module (NLI), which uses a homemade RNLI dataset for migration learning (model fine-tuning) on a pre-trained model and stores the inferred sentence relations and relational degrees in a database for matching with user input sentences; 2) an inference library matching (DSM) module, which uses a sentence matching algorithm to match the input accident statements with the The matching result is the correspondence of the input accident text. Based on the results of the two modules, the system is given a comprehensive score, and after several experiments the system is able to achieve an accuracy rate of 79.7% for textual event inference, which verifies the effectiveness of the system.The system not only handles text incidents efficiently, objectively and accurately, the reasoning results can also be used as a safety guide for equipment personnel and technicians, helping the relevant areas to develop appropriate daily work lists and eliminate safety hazards.
|
