We're sunsetting PodQuest on 2025-07-28. Thank you for your support!
Export Podcast Subscriptions
cover of episode #113: Why you need Knowledge Graphs for your AI chatbot | ft. Aniket Mitra

#113: Why you need Knowledge Graphs for your AI chatbot | ft. Aniket Mitra

2025/2/5
logo of podcast Real World Serverless with theburningmonk

Real World Serverless with theburningmonk

AI Deep Dive AI Chapters Transcript
People
A
Aniket Mitra
Topics
Aniket Mitra: 我是Saveway公司的创始人,我们专注于AI、ML和数据工程项目,尤其擅长构建企业级知识图谱。我曾为地图和零售领域的客户构建知识图谱,目前正在进行一些结合知识图谱和LLM的项目。知识图谱通过赋予数据语义意义,将信息转化为知识,其结构天然呈图状。这得益于本体论的概念,本体论定义了属性的语义,使知识图谱能够推断出未明确记录的信息。图数据库(如Neo4j)优于关系数据库,因为它能建模语义,捕获隐含信息,并支持属性和关系属性的查询;但知识图谱不应是万能的,应与其他系统(如Elasticsearch)结合使用。结合向量数据库和图数据库可以更好地利用LLM:向量数据库捕捉基于位置的关联,图数据库捕捉语义相似性,两者结合可以获得更佳效果,甚至可以结合图神经网络生成嵌入。在LLM增强检索中,图数据库和向量数据库协同工作:图数据库用于处理实体关系和语义关系,向量数据库用于处理语义相似性;可以采用多Agent系统或结合两种嵌入的方式来整合信息。LLM的出现正在解决知识图谱构建中的两个核心问题:一是利用LLM实现本体论的半自动化管理,二是利用LLM改进查询解释,减少对人工干预的依赖。知识图谱可以用于改进RAG,特别是针对特定领域:它可以帮助选择用于微调过程的文档和信息,并通过结构化信息来减少上下文长度的限制,从而提高微调效率。这形成了一个闭环:微调后的LLM可以更好地将非结构化数据转换为结构化数据,从而改进知识图谱。我的团队正在研究如何利用LLM更高效地构建和使用知识图谱,欢迎感兴趣的人联系我合作。 Jan: (问题和引导性发言,未形成核心论点)

Deep Dive

Chapters
This chapter introduces knowledge graphs, their history, and their resurgence in the age of AI and LLMs. It explains how knowledge graphs organize data semantically, adding meaning beyond traditional relational databases, and forming a graph structure that facilitates knowledge representation.
  • Knowledge graphs organize data semantically, adding meaning beyond traditional relational databases.
  • The structure of knowledge in a knowledge graph is in the form of a graph.
  • Ontologies assign meaning to attributes in the metadata space, transforming information into knowledge.

Shownotes Transcript

In this episode, I spoke with Aniket Mitra, founder of CW, to learn more about knowledge graphs and how they are making a comeback in the era of AI and LLMs.Opening theme song:Cheery Monday by Kevin MacLeodLink: https://incompetech.filmmusic.io/song/3495-cheery-mondayLicense: http://creativecommons.org/licenses/by/4.)