图模互补:知识图谱与大模型融合综述
KG-LLM-MCom:A Survey on Integration of Knowledge Graph and Large Language Model
黄勃 1吴申奥 1王文广 2杨勇 3刘进 4张振华 1陈南希 5杨洪山6
作者信息
- 1. 上海工程技术大学电子电气工程学院,上海 201620
- 2. 达观数据有限公司,上海 201203
- 3. 上海宝信软件股份有限公司,上海 201203
- 4. 武汉大学计算机学院,湖北武汉 430072
- 5. 中国科学院上海微系统与信息技术研究所,上海 200050
- 6. 星环信息科技(上海)股份有限公司,上海 200233
- 折叠
摘要
大模型(LLM)的兴起在自然语言处理领域引起了广泛关注,其涌现能力在各个垂直领域(如金融、医疗、教育等)也取得一定进展.然而,大模型自身面临解释性不足、知识实时性差、生成结果存在虚假信息等诸多挑战.为了应对这些问题,知识图谱与大模型的融合逐渐成为了研究热点.知识图谱作为一种结构化的知识模型,其真实性和可靠性,成为提高大模型解释和推理能力的有力工具.同时大模型具备语义理解能力,为知识图谱的构建和更新提供了有力支持.因此,知识图谱和大模型是互补的(本文称为图模互补).本文系统性地介绍知识图谱与大模型融合的方法,分别从1)大模型增强知识图谱,2)知识图谱增强大模型两个角度进行全面回顾和分析.最后,本文从医学诊断预测和时间知识图谱出发,介绍图模互补的领域应用,并讨论图模互补未来发展的方向,为知识图谱与大模型的进一步研究提供帮助.
Abstract
The rise of the Large Language Model(LLM)has attracted wide attention in Natural Language Processing,and its emergent ability has also made some progress in various vertical fields(such as finance,healthcare,education,etc.).However,LLM faces numerous challenges,including inadequate interpretation,limited real-time knowledge,and the potential presence of false information in generated outcomes.In order to address these problems,integrating Knowledge Graphs(KG)and LLM has gradually become a research hotspot.As a structured knowledge model,KG has become a powerful tool for improving the interpre-tation and reasoning ability of LLM due to its authenticity and reliability.At the same time,LLM has the ability to understand se-mantics,which supports the construction and update of knowledge graphs.Therefore,KG and LLM are mutually complementary(KG-LLM-MCom is called in this article).The article provides a systematic introduction to KG and LLM integration methods.It conducts a comprehensive review and analysis from two perspectives:1)LLM-enhanced KG and 2)KG-enhanced LLM.Finally,the article introduces the field applications of KG-LLM-MCom from the viewpoint of Medical Diagnosis prediction and Temporal Knowledge Graphs.It discusses KG-LLM-MCom's future development direction,which can provide help for further research on KG and LLM.
关键词
大模型/知识图谱/自然语言处理Key words
large language model/knowledge graph/natural language processing引用本文复制引用
基金项目
科技部科技创新2030"新一代人工智能"重大项目(2020AAA0109300)
上海地方能力建设项目(21010501500)
上海市科技创新行动计划(21DZ1204900)
出版年
2024
NETL
NSTL
万方数据