知识图谱——提升油气行业大模型RAG 性能的关键技术

doi:10.3969/j.issn.1002-302X.2024.06.014

摘要/Abstract

摘要： 针对传统RAG在关联分析、信息整合与逻辑推理能力等方面存在的局限性，以知识图谱与RAG为研究对象，分析国内外研究进展与应用案例。ChatLaw引入领域专家精准定义法律实体、关系与案例，通过高质量知识图谱提升法律咨询的准确性。GraphRAG采用知识图谱表示非结构文本中实体与关系，通过层次聚类、摘要生成等技术提升RAG在大规模数据集上全局搜索能力。HippoRAG在查询阶段利用知识图谱进行概念扩展和检索，提升RAG知识整合与多跳推理能力。归纳RAG与知识图谱融合方法，在数据分块、数据存储、查询优化、检索召回、重排、提示词构建、答案生成等阶段引入知识图谱，可以提升RAG准确率、关联分析能力、推理能力与可解释性。基于Lucene、LangChain等开源框架设计全文检索、向量检索、图谱检索3 套方案，将其应用于油气知识问答场景，验证知识图谱对增强RAG 的有效性。

关键词: 大语言模型, 检索增强生成, 知识图谱, 向量数据库, 图数据库

Abstract: Addressing the limitations of traditional Rule-Augmented Generation (RAG) in aspects such as associative analysis, information integration, and logical reasoning capabilities, this study focuses on the integration methods of knowledge graphs and RAG by analyzing the research progress and application cases at home and abroad. ChatLaw introduces domain experts to precisely define legal entities, relationships, and cases, enhancing the accuracy of legal consultations through high-quality knowledge graphs. GraphRAG employs knowledge graphs to represent entities and relationships in unstructured text, leveraging techniques such as hierarchical clustering and summarization generation to improve RAG’s global search capabilities on large-scale datasets. HippoRAG utilizes knowledge graphs for concept expansion and retrieval during the query step, enhancing RAG’s knowledge integration and multi-hop reasoning abilities. This study summarizes fusion methods of RAG and knowledge graphs, suggesting that incorporating knowledge graphs in stages such as data partitioning, data storage, query optimization, retrieval recall, reranking, prompt construction,and answer generation can improve RAG’s accuracy, associative analysis capabilities, reasoning abilities, and interpretability. Based on open-source frameworks such as Lucene and LangChain, three retrieval schemes—full-text retrieval, vector retrieval, and graph retrieval—are designed and applied to oil and gas knowledge Q&A scenarios, validating the effectiveness of knowledge graphs in enhancing RAG.

Key words: large language models, retrieval-augmented generation, knowledge graph, vector database, graph database

中图分类号:

TP391
TP18

宋子瑜. 知识图谱——提升油气行业大模型RAG 性能的关键技术[J]. 石油科技论坛, 2024, 43(6): 114-125.

Song Ziyu. Knowledge Graphs: Key Technology for Enhancing RAG Performance in the Oil and Gas Industry[J]. Petroleum Science and Technology Forum, 2024, 43(6): 114-125.

参考文献

[1] Zhao Wayne Xin, Zhou Kun, Li Junyi, et al. A survey of large language models[EB/OL]. [2024-06-12]. https://arxiv.org/abs/2303.18223v15.
[2] Lewis P, Perez E, Piktus A, et al. Retrieval-augmented generation for knowledge-intensive NLP tasks[EB/OL]. [2024-06-12]. https://arxiv.org/abs/2005.11401?file=2005.11401.
[3] Gao Yunfan, et al. Retrieval-augmented generation for large language models: A survey[EB/OL]. [2024-06-12]. https://arxiv. org/abs/2312.10997.
[4] 赵鑫, 窦志成, 文继荣. 大语言模型时代下的信息检索研究发展趋势[J]. 中国科学基金, 2023, 37(5): 786-792.
Zhao Xin, Dou Zhicheng, Wen Jirong. The development of information retrieval in the era of large language model[J]. Bulletin of National Natural Science Foundation of China, 2023, 37(5): 786-792.
[5] 车万翔, 窦志成, 冯岩松, 等. 大模型时代的自然语言处理: 挑战、机遇与发展[J]. 中国科学(信息科学), 2023, 53(9): 1645-1687.
Che Wanxiang, Dou Zhicheng, Feng Yansong, et al. Towards a comprehensive understanding of the impact of large language models on natural language processing: Challenges, opportunities and future directions[J]. Science in China (Information Sciences), 2023, 53(9):1645-1687.
[6] Yikun Han, Chunjiang Liu, Pengfei Wang. A comprehensive survey on vector database: Storage and retrieval technique, challenge[EB/OL].[2024-06-12]. https://arxiv.org/abs/2310.11703.
[7] 张浩. 基于向量数据库的智能媒资搜索研究[J]. 电声技术, 2022, 46(1): 22-24, 28.
Zhang Hao. Research on intelligent media resources search based on vector database[J]. Audio Engineering, 2022, 46(1): 22-24, 28.
[8] Ji Shaoxiong, Pan Shirui, Erik Cambria, et al. A survey on knowledge graphs: Representation, acquisition, and applications[J]. IEEE Transactions on Neural Networks and Learning Systems, 2022, 33(2):494-514.
[9] 刘世侠, 李卫军, 刘雪洋, 等. 基于强化学习的知识图谱推理研究综述[J]. 计算机应用研究, 2024, 41(9): 2561-2572.
Liu Shixia, Li Weijun, Liu Xueyang, et al. Review of reinforcement learning based knowledge graph reasoning research[J]. Application Research of Computers, 2024, 41(9): 2561-2572.
[10] Pan Shirui, Luo Linhao, Wang Yufei, et al. Unifying large language models and knowledge graphs: A roadmap[EB/OL]. [2024-06-12].https://arxiv.org/pdf/2306.08302v1.
[11] 唐晓晟, 程琳雅, 张春红, 等. 大语言模型在学科知识图谱自动化构建上的应用[J]. 北京邮电大学学报( 社会科学版), 2024, 26(1):125-136.
Tang Xiaosheng, Cheng Linya, Zhang Chunhong, et al. Application of large language models in automated construction of knowledge graphsfor university subject domains[J]. Journal of Beijing University of Posts and Telecommunications (Social Sciences Edition), 2024, 26(1):125-136.
[12] 余燕芳, 夏亮亮, 李翼鸿, 等. 基于知识图谱和大语言模型的终身学习资源库供给生态构建研究[J]. 远程教育杂志, 2024, 42(1): 104-112.
Yu Yanfang, Xia Liangliang, Li Jihong, et al. Research on the supply ecology of lifelong learning resource database based on knowledge graph and large language model[J]. Journal of Distance Education,2024, 42(1): 104-112.
[13] 王智悦, 于清, 王楠, 等. 基于知识图谱的智能问答研究综述[J]. 计算机工程与应用, 2020, 56(23): 1-11.
Wang Zhiyue, Yu Qing, Wang Nan, et al. Survey of intelligent question answering research based on knowledge graph[J]. Computer Engineering and Applications, 2020, 56(23): 1-11.
[14] 文森, 钱力, 胡懋地, 等. 基于大语言模型的问答技术研究进展综述[J]. 数据分析与知识发现, 2024, 8(6):16-29.
Wen Sen, Qian Li, Hu Maodi, et al. Review of research progress on question-answering techniques based on large Language models[J].Data Analysis and Knowledge Discovery, 2024, 8(6): 16-29.
[15] 官璐, 何康, 斗维红. 微调大模型: 个性化人机信息交互模式分析[J].新闻界, 2023(11): 44-51, 76.
Guan Lu, He Kang, Dou Weihong. Fine-tuning large language models:The analysis of personalized human-computer information interaction patterns[J]. Journalism and Mass Communication, 2023(11): 44-51, 76.
[16] 孙维纬. 知识检索增强的对话系统研究[D]. 济南: 山东大学, 2023.Sun Weiwei. Research on knowledge retrieval augmented dialogue systems[D]. Jinan: Shandong University, 2023.
[17] Cui Jiaxi, Ning Munan, Li Zongjian, et al. Chatlaw: A multi-agent collaborative legal assistant with knowledge graph enhanced mixtureof-experts large language model[EB/OL]. [2024-06-12]. https://arxiv.org/pdf/2306.16092.
[18] Xu Zhentao, Cruz M J, Guevara M, et al. Retrieval-augmented generation with knowledge graphs for customer service question answering[EB/OL]. [2024-06-12]. https://arxiv.org/abs/2404.17723.
[19] Darren Edge, Ha Trinh, Newman Cheng, et al. From local to global: A graph RAG approach to query-focused summarization[EB/OL]. [2024-06-12]. https://arxiv.org/abs/2404.16130.
[20] Jiang Xinke, Zhang Ruizhe, Xu Yongxin, et al. HyKGE: A hypothesis knowledge graph enhanced framework for accurate and reliable medical LLMs responses[EB/OL]. [2024-06-12]. https://arxiv. org/abs/2312.15883.
[21] Gutiérrez B J, Shu Yiheng , Gu Yu, et al. HippoRAG: Neurobiologically inspired long-term memory for large language models[EB/OL]. [2024-06-12]. https://arxiv.org/pdf/2405.14831.