Agentic Design

Patterns
🕸️

Graph RAG(GRAG)

Knowledge graph-enhanced RAG using entity relationships and community detection for global sensemaking queries

Complexity: highKnowledge Retrieval (RAG)

🎯 30-Second Overview

Pattern: Knowledge graph-based RAG with community detection and hierarchical summarization for large-scale reasoning

Why: Enables complex multi-hop reasoning and relationship understanding through structured graph traversal

Key Insight: Microsoft's approach uses community detection algorithms to create hierarchical summaries for global and local reasoning

⚡ Quick Implementation

1Extract & Build:NER/RE to extract entities and relations, build knowledge graph
2Index & Store:Create graph indexes and hybrid text-vector search capabilities
3Query & Map:Parse query entities and map to graph nodes
4Traverse & Retrieve:Multi-hop graph traversal with bounded expansion
5Assemble & Generate:Serialize graph context and generate response
Example: query → entity_linking → graph_traversal → subgraph_assembly → llm_generation → response

📋 Do's & Don'ts

✅Implement community detection algorithms for large-scale graph summarization
✅Use hierarchical indexing with global and local community summaries
✅Apply entity resolution and deduplication to maintain graph quality
✅Combine vector search with graph traversal for hybrid retrieval
✅Implement bounded traversal with explicit hop limits (1-3 hops)
❌Allow uncontrolled graph traversal that leads to context explosion
❌Skip entity linking validation and confidence scoring
❌Neglect graph maintenance and consistency validation
❌Create overly complex schemas that hinder query performance
❌Ignore provenance tracking for graph edges and entity sources

🚦 When to Use

Use When

  • • Complex multi-hop reasoning requiring entity relationship understanding
  • • Knowledge-intensive domains with rich interconnected information
  • • Enterprise data with well-defined entity schemas and relationships
  • • Fact-checking and verification requiring structured evidence paths
  • • Large-scale knowledge bases needing hierarchical summarization

Avoid When

  • • Simple factual queries adequately served by document-based RAG
  • • Domains with poor entity extraction and relation modeling quality
  • • Real-time applications with strict latency requirements
  • • Small datasets where graph complexity exceeds retrieval benefits
  • • Applications lacking well-defined entity schemas or ontologies

📊 Key Metrics

Graph Construction Quality
Precision/recall of extracted entities and relations
Community Detection Accuracy
Coherence and relevance of detected communities
Retrieval Relevance
Proportion of relevant subgraphs in query results
Multi-hop Reasoning
Accuracy of complex relationship inference
Context Completeness
Coverage of relevant graph neighborhoods
Query Response Time
End-to-end latency including graph operations

💡 Top Use Cases

Biomedical Research: Drug-disease-gene relationship exploration with scientific literature integration
Financial Analysis: Company-industry-market relationship modeling for investment research
Legal Research: Case law relationship mapping with precedent and citation analysis
Enterprise Knowledge: Organizational relationship modeling with department and project connections
Academic Research: Citation networks and research collaboration graph analysis

References & Further Reading

Deepen your understanding with these curated resources

Foundational Papers & Microsoft GraphRAG

From Local to Global: A Graph RAG Approach to Query-Focused Summarization (Edge et al., 2024)

GraphRAG: Unlocking LLM Discovery on Narrative Private Data (Microsoft Research, 2024)

Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (Lewis et al., 2020)

Knowledge Graph-Enhanced Large Language Models via Path Selection (Wang et al., 2023)

Community Detection & Graph Algorithms

Leiden Algorithm for Community Detection in Large Networks (Traag et al., 2019)

Fast Unfolding of Communities in Large Networks (Blondel et al., 2008)

Graph Neural Networks: A Review of Methods and Applications (Zhou et al., 2020)

Network Analysis and Graph Theory in Python (NetworkX Documentation)

Knowledge Graph Construction & NER/RE

REBEL: Relation Extraction By End-to-end Language generation (Huguet Cabot & Navigli, 2021)

spaCy: Industrial-Strength Natural Language Processing

OpenIE: Open Information Extraction (Banko et al., 2007)

Stanford CoreNLP: A Java suite of core NLP tools

Graph Database & Storage Solutions

Neo4j Graph Database: Developer Guide and Best Practices

ArangoDB Multi-Model Database with Graph Capabilities

Amazon Neptune: Fully Managed Graph Database Service

TigerGraph: Scalable Graph Analytics Platform

Implementation Frameworks & Tools

Microsoft GraphRAG Official Implementation and Documentation

LangChain Graph RAG Integration and Examples

LlamaIndex Knowledge Graph Implementation

Neo4j Vector Search and Graph RAG Tutorial

Evaluation & Benchmarking

HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering

ComplexWebQuestions: Multi-hop Question Answering over Knowledge Graphs

KGQA: Knowledge Graph Question Answering Benchmark

Graph-based RAG Evaluation Metrics and Frameworks

Contribute to this collection

Know a great resource? Submit a pull request to add it.

Contribute

Patterns

closed

Loading...

Built by Kortexya