Enhancing Named Entity Recognition in Safety Hazard Analysis through GBD and LLMs
Haoling Dong, Bin Ying Wu · 2024
With the continuous enhancement of informatization in production safety, the need to strengthen the analysis capability of big data in production safety is increasingly growing. This is crucial for preventing major accidents and ensuring production safety. However, the main challenge currently faced is that information about safety hazards often appears in unstructured text formats, lacking a unified standard and normative system. This makes it difficult to standardize knowledge extraction from hazard information. Additionally, traditional named entity recognition methods arc inadequate in understanding sentence structures, lexical components, and the dependency relationships between words, especially in processing texts in specific complex domains, where their entity recognition effects arc still insufficient. Given this, the research focus of this paper includes: deeply exploring the concept and connotation of big data in hazard information, clarifying the classification system of hazard information; designing a standard hazard information annotation template specifically for the field of production safety; designing a named entity recognition method based on GBD; designing and implementing a named entity recognition method based on LLMs; and based on the recognition results, constructing a knowledge graph of the relevant field to promote its effective utilization in practical applications.