Scholarlink

In the research field of rockburst prevention and control, there are problems of complicated knowledge and difficulty in obtaining. People in the coal mine field often spend a lot of energy learning related knowledge. In addition, there is a serious hallucination problem in the professional field of large language models. To disseminate rockburst-related knowledge and address its fragmented nature, this work proposes RB-RAG (Rockburst Retrieval-Augmented Generation) based on the RAG(Retrieval-Augmented Generation) method, taking rockburst knowledge as a data set and taking it as a part of Prompt strategy based on CoT (Chain-of-Thought). Various metrics are used to evaluate the proposed RB-RAG, among which the highest faithfulness is 0.9481. Results show that our rockburst knowledge intelligent system has the characteristics of improving the retrieval speed and reducing the hallucination in the generation of large language models. This advancement establishes a foundation for widespread dissemination of rockburst knowledge.

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

RB-RAG: Intelligent Question Answering System of Rock Burst Knowledge Based on Retrieval-Augmented Generation

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Kangning Bu,
  • Zehua Wang,
  • Pengcheng Zhu,
  • Parviz Isaev,
  • Wei Chen,
  • Tingting Xu,
  • Jueting Liu

摘要

In the research field of rockburst prevention and control, there are problems of complicated knowledge and difficulty in obtaining. People in the coal mine field often spend a lot of energy learning related knowledge. In addition, there is a serious hallucination problem in the professional field of large language models. To disseminate rockburst-related knowledge and address its fragmented nature, this work proposes RB-RAG (Rockburst Retrieval-Augmented Generation) based on the RAG(Retrieval-Augmented Generation) method, taking rockburst knowledge as a data set and taking it as a part of Prompt strategy based on CoT (Chain-of-Thought). Various metrics are used to evaluate the proposed RB-RAG, among which the highest faithfulness is 0.9481. Results show that our rockburst knowledge intelligent system has the characteristics of improving the retrieval speed and reducing the hallucination in the generation of large language models. This advancement establishes a foundation for widespread dissemination of rockburst knowledge.