Making Your Knowledge Graph LLM-Ready: Quality Assessment and Design Strategies for GraphRAG
summary
GraphRAG uses either text-to-query → schema-derived Cypher/SPARQL/SQL or fact-embeddings → vector DB retrieval and succeeds only if retrieval, generation and the KG itself are high quality. Measure with end-to-end answer correctness plus relevance, faithfulness, and contextual precision/recall to isolate failures. Watch for six KG issues: plain false facts (e.g., “Greece capital = New York”), contradictions, missing facts, redundancy/duplicates, irrelevant domain bleed, and unreadable labels. Detect with logical constraints, heuristics (e.g., movies with <3 actors), or ML/LLM detectors.
AI-written summary · about
chapters (6)
- 0:00Introduction and overview
- 3:37What is GraphRAG
- 8:37Assessing GraphRAG quality
- 14:39Knowledge graph quality issues
- 22:01Detecting knowledge graph errors
- 25:12Key takeaways and conclusion
related talks
20mPractical GraphRAG: Making LLMs smarter with Knowledge Graphs
53mPractical GraphRAG - Making LLMs smarter with Knowledge Graphs
26mPractical GraphRAG Making LLMs smarter with Knowledge Graphs
46mKnowledge Graphs + LLMs: What's the latest in research?
27mLies, Damn Lies, and AIs
23mKnowledge Graphs and LLMs in Action
52mGive Your LLMs a Left Brain
43mIncreasing the LLM Accuracy for Question Answering on Structured Data: Knowledge Graphs to the Rescu