In the evolving landscape of AI and information retrieval, knowledge graphs have emerged as a powerful way to represent complex, interconnected information. But how do they work, and are they better than traditional RAG setups?
Resources:
Notebook Repo - https://github.com/ALucek/GraphRAG-Breakdown
GraphRAG - https://microsoft.github.io/graphrag/
GraphRAG Paper - https://arxiv.org/pdf/2404.16130
GraphRAG Case Study from LinkedIN - https://arxiv.org/pdf/2404.17723
Unifying Large Language Models and Knowledge Graphs - https://arxiv.org/pdf/2306.08302
The Ultimate Guide to Fine-Tuning LLMs - https://arxiv.org/pdf/2408.13296
Google’s Knowledge Graph Introduction - https://blog.google/products/search/introducing-knowledge-graph-things-not/
Leiden Algorithm - https://en.wikipedia.org/wiki/Leiden_algorithm
Chapters:
00:00 - Why RAG Fails
01:54 - What is a Knowledge Graph?
03:35 - Knowledge Graphs & LLMs
05:39 - Introducing GraphRAG
06:17 - Main Components of Knowledge Graphs
07:39 - Setting up GraphRAG
11:10 - Data Flow: Overview
12:13 - Data Flow: Entity & Relationship Extraction
16:40 - Data Flow: Community Clustering
18:28 - Data Flow: Community Report Generation
20:11 - Observing Final Knowledge Graph
22:12 - RAG Setup
23:58 - RAG: Local Search
27:10 - RAG: Global Search
30:23 - RAG: DRIFT Search
35:06 - Comparing GraphRAG vs Regular RAG
36:59 - Comparison Discussion
#ai #datascience #programming