Scholarlink

<p>The effective prediction and management of potential hazards are critical to ensuring safety in mining operations. With advancements in generative artificial intelligence (AI) and natural language processing (NLP), there is an opportunity to automate the analysis of mining safety data. A novel summarization model, that took advantage of domain expertise, was developed to work with OpenAI’s GPT-3.5-turbo Large Language Model (LLM) and applied to Mine Safety and Health Administration (MSHA) data. When asked to summarize narratives, the model first predicts different attributes of mine safety from individual narratives. It then summarizes the predicted attributes, while actively suppressing predicted attributes that are deemed unreliable. Therefore, the resultant verbose summary is not only informative but also accurate. Domain knowledge was leveraged at every stage of model development, from the selection of attributes for prediction to the summarization method. The model’s performance was evaluated using standard metrics such as the F1 score, showing an accuracy of 86%. The achieved performance was vastly superior when compared to the default LLM summarization. This work highlights the benefits of combining domain expertise with AI technologies to enhance hazard identification in the mining industry.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Predictive Summary Model—a Domain-Guided Approach to Generate Informative Summaries

  • 期刊论文

  • 发表日期：

    2025年8月13日

查看全文
  • Raviteja Tatikonda,
  • Rajive Ganguli

摘要

The effective prediction and management of potential hazards are critical to ensuring safety in mining operations. With advancements in generative artificial intelligence (AI) and natural language processing (NLP), there is an opportunity to automate the analysis of mining safety data. A novel summarization model, that took advantage of domain expertise, was developed to work with OpenAI’s GPT-3.5-turbo Large Language Model (LLM) and applied to Mine Safety and Health Administration (MSHA) data. When asked to summarize narratives, the model first predicts different attributes of mine safety from individual narratives. It then summarizes the predicted attributes, while actively suppressing predicted attributes that are deemed unreliable. Therefore, the resultant verbose summary is not only informative but also accurate. Domain knowledge was leveraged at every stage of model development, from the selection of attributes for prediction to the summarization method. The model’s performance was evaluated using standard metrics such as the F1 score, showing an accuracy of 86%. The achieved performance was vastly superior when compared to the default LLM summarization. This work highlights the benefits of combining domain expertise with AI technologies to enhance hazard identification in the mining industry.