Knowledge graph based chatbots
summary
Describes replacing blind LLM output with a knowledge graph backed flow: build a graph of entities and relations, generate Cypher query generation from questions, run queries to fetch exact facts, and feed those results plus relevant documents to ChatGPT to avoid invented claims. Shows vector pipelines for retrieval, tradeoffs like preprocessing cost and duplicate hits, and demos answering multihop queries and summarizing news. Centers on explicit control over answers and vector similarity search as concrete tools.
AI-written summary · about
chapters (10)
- 0:00Knowledge graph chatbots intro
- 1:06LLM chatbot limitations
- 4:17Unstructured data for better answers
- 6:43Vector embeddings for smart search
- 10:13Multi-hop questions solved
- 13:11Structured + unstructured data
- 18:55Live demo
- 23:16Benefits and accessibility
- 24:23Summary
- 25:48Q&A
related talks
28mBuilding Knowledge Graph-Based Agents with Structured Text Generation
19mWisdom-Driven Knowledge Augmented Generation at Scale
51mA ChatGPT on your own data: how does it work?
32mMulti-Agent Query Generation w/ LLMs on Complex Knowledge Graphs
25mEnhance LLM Applications' Trustworthiness and Explainability
31mLive Q&A
14mKnowledge Graphs as the Backbone of AI-First Software Delivery
23mKnowledge Graphs and LLMs in Action