Knowledge graphs and RAG
Combining **knowledge graphs** (entities + relations) with RAG so retrieval can follow structured links—not only vector similarity.
What it is
Combining knowledge graphs (entities + relations) with RAG so retrieval can follow structured links—not only vector similarity.
Why it matters
Some questions are relational (“who reports to whom?”). Graphs help; they also need curation discipline.
How it works (plain)
Extract entities/relations → store graph → at query time traverse + fetch text chunks → generate with citations to both.
Try it
Draw a 5-node graph for a domain you know; write one question that needs a hop.
Myths
- ⚠️ Myth: Graphs remove hallucinations automatically.
- ✓ Reality: Bad extractions edges become confident wrong answers.
Sources
- Course 09 RAG; Course 06 logic; Course 15 NER
- Course 02 provenance