.png){kind=icon}
.png){kind=icon}
Return
Implementing generative pretrained transformer models for text recognition tasks in safety data sheets
.png){kind=icon}
OA
AI
DOI：10.1093/annweh/wxaf081.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Workplaces handling chemicals require an up-to-date and comprehensive assessment of the potential risks for their workforce. Online safety data sheets (SDSs) inventories provide adequate information to perform risk assessments. However, current practices that manually import information from SDSs into the online inventories are time-consuming, leading to delayed or inadequate risk assessments. This study presents a pipeline using large language models (LLMs) to automate the extraction and management of data from SDSs to online chemical inventories. The pipeline achieved an average accuracy of 0.83 in (close to precisely) extracting multiple variables of interest, such as company name, product name, and hazard statements, in comparison to manually extracting these variables. Overall, this pipeline illustrates the ability of LLM tools to automate SDS inventory management and thereby support the possibility to perform up-to-date risk assessments and evaluation tasks on the work floor, ultimately contributing to occupational safety.
Keywords:
large language models
occupational safety
safety data sheets
text extraction
Journal
IF:
2.1
Papers:
191
Citations:
1.5K