哈尔滨商业大学学报（自然科学版）2025，Vol.41Issue(6)：661-667,706,8.
基于互信息的领域问答检索增强生成
Mutual information-based retrieval-augmented generation(RAG)for domain question answering
卢海涛 1郑大宇 1杨跞 2常柏灵 1于信鑫1
作者信息
- 1. 哈尔滨商业大学 轻工学院,哈尔滨 150028
- 2. 上海新松机器人有限公司,上海 201304
- 折叠
摘要
Abstract
The Retrieval-Augmented Generation(RAG)reduced hallucinations in large language models(LLM)for domain-specific question and answer,but its retrieval-first approach could yield poor results if the knowledge base or retrieval was flawed.Therefore,a mutual information-based retrieval-augmented generation(MI-RAG)method for domain Q&A was proposed.The knowledge graphs(KG),LLM and external network knowledge were used as data sources by MI-RAG.In the retrieval process,mutual information(MI)was used as the retrieval index,and LLM and graph retrieval were used to retrieve the subgraph most relevant to the problem from KG.After generating answers,LLM was also used to update the KG.MI-RAG was evaluated on the domain Q&A task and showed significant superiority over existing state-of-the-art models and other retrieval-augmented generation methods.关键词
大语言模型/知识图谱/检索增强生成/提示词工程/互信息/领域问答Key words
large language models/knowledge graphs/retrieval-augmented generation/prompt engineering/mutual information/domain question answering分类
信息技术与安全科学引用本文复制引用
卢海涛,郑大宇,杨跞,常柏灵,于信鑫..基于互信息的领域问答检索增强生成[J].哈尔滨商业大学学报（自然科学版）,2025,41(6):661-667,706,8.
