Adaptive GraphRAG: A Framework for Knowledge Graph Quality, Consistency, Evolution
summary
Adaptive GraphRAG is a Neo4j-based lifecycle for keeping generated knowledge graphs accurate and traceable as new text arrives, using Cypher, GDS, vector indexes and LLMs as controlled aides. It implements repeatable operators like dedupe operator (vector-nearest + name normalization + APOC merge), prediction operator (Adamic‑Adar + ontology + vector score), and aggregation to build core/child hierarchies. Health metrics include giant-component growth, duplicate density and evidence coverage. A diabetes companion prototype runs on Neo4j Aura and demonstrates improved connectivity and provenance over naive extraction.
AI-written summary · about
chapters (14)
- 0:00Introduction and motivation
- 0:51Problem with static knowledge graphs
- 3:21Generative knowledge graph lifecycle
- 5:27Adaptive GraphRAG operational loop
- 6:49Measuring graph quality
- 8:35Generative knowledge modeling
- 9:19Document graph and multi-layer structure
- 12:13Ontology graph for semantic stability
- 15:47Operators for graph refinement
- 17:10Deduplication operator details
- 19:17Prediction operator for implicit links
- 20:45Aggregation operator for hierarchy
- 22:15Case study diabetes companion system
- 23:42Conclusion and future outlook
related talks
39mEnhancing RAG with Neo4j Knowledge Graph
25mGraphRAG: Using LLMs to Build and Interrogate a Regulatory Knowledge Graph
16mEnhancing Retrieval-Augmented Generation with GraphRAG Patterns in Neo4j
25mEnhancing Job Matching With Knowledge Graphs and RAG
30mAgentic RAG with Knowledge Graphs Explained
33mSmarter Industrial Fault Diagnosis with Knowledge Graphs and Large Language Models
30mPerformant GraphRAG Improving Neo4j to Drive Autonomous AI ...
46mBuilding Knowledge Graphs