PRODUCT UPDATE

Discover Hidden Patterns: Introducing Graph Communities

Automatically detect and visualize communities in your knowledge graphs. Search for related concepts using natural language.

The Challenge: Lost in Complexity

Knowledge graphs grow quickly. What starts as a handful of entities and relationships can balloon into hundreds or thousands of nodes. As your graph expands, understanding its structure becomes increasingly difficult:

• Where do I start? — With hundreds of nodes, finding relevant information feels like searching for a needle in a haystack
• What's related to what? — Relationships exist, but the bigger picture remains unclear
• How do I navigate? — Manual exploration is time-consuming and inefficient

Traditional graph visualization shows you every node and edge, but doesn't help you understand the underlying structure or find what you're looking for quickly.

The Solution: Community Detection

Our new Graph Communities feature automatically identifies clusters of related entities in your knowledge graph using advanced graph clustering algorithms. These communities represent natural groupings — topics, themes, or domains that emerge from your data.

What Are Communities?

Communities are groups of nodes that are more densely connected to each other than to the rest of the graph. Think of them as:

• Topic clusters — All entities related to "data privacy" grouped together
• Domain areas — Legal concepts, technical specifications, or business processes organized naturally
• Conceptual neighborhoods — Related ideas that frequently appear together in your documents

How It Works

1. Automatic Detection — Our Leiden clustering algorithm analyzes your graph structure and identifies communities without any manual configuration
2. Smart Naming — An LLM generates human-readable names and descriptions for each community based on the entities it contains
3. Visual Representation — Communities are displayed with colored bounding boxes, making the graph structure immediately understandable
4. Natural Language Search — Find communities using conversational queries like "show me the data privacy community" or "what communities exist?"

Key Features

Visual Community Boundaries

Communities are displayed with semi-transparent colored borders around related nodes. Each community has:

• Unique color — Consistent color assignment for easy identification
• Bounding box — Visual boundary showing all nodes in the community
• Community name — LLM-generated title displayed above the bounding box
• Interactive elements — Click to highlight, hover for details

Natural Language Community Search

Use the chat interface to find and explore communities:

User: "Show me the Data Privacy community"
AI: "Found 1 community matching 'Data Privacy': Data Privacy Community (12 nodes)"
AI: "Highlighting the community in the graph."

User: "What communities exist?"
AI: "I found 5 communities in your graph:
1. Data Privacy Community (12 nodes)
2. Machine Learning Research (8 nodes)
3. Legal Cases (5 nodes)
4. Company Structure (7 nodes)
5. Product Features (6 nodes)"

Community Summaries

Each community includes an LLM-generated summary that explains:

• What the community represents
• Key entities and concepts
• How the community relates to your overall knowledge base

Important Note on Visualization

Community visualizations appear during search interactions. When you search for communities using the chat interface, the graph will highlight and display community boundaries. On initial graph load, you'll see the standard node-and-edge visualization without community overlays. This design ensures optimal performance while providing community insights when you need them.

Use Cases

Knowledge Base Exploration

New team members can quickly understand your knowledge base structure by exploring communities. Instead of navigating hundreds of individual nodes, they can start with high-level community groupings and drill down as needed.

Cross-Document Insights

Communities often span multiple documents, revealing connections that aren't obvious from individual files. A "compliance" community might include entities from legal documents, policy files, and technical specifications — showing the full scope of your compliance landscape.

Targeted Research

When researching a specific topic, find the relevant community and focus your exploration there. This is especially useful for large knowledge bases where manual searching would be impractical.

Graph Health Monitoring

Track how communities evolve as you add new documents. New communities may emerge, existing ones may grow or merge, giving you insights into how your knowledge base is developing.

Getting Started

Graph Communities is available now for all accounts using GraphRAG with knowledge graphs.

To use community features:

1. Open your knowledge graph — Navigate to any RAG project with an existing knowledge graph
2. Use the chat interface — Ask questions like:
   • "What communities exist in this graph?"
   • "Show me the [topic] community"
   • "Highlight communities related to [concept]"
3. Explore visually — Click on community boundaries to highlight all nodes in that community
4. View summaries — Hover over community titles to see LLM-generated descriptions

Technical Details

For developers and technical users, here's how community detection works:

• Algorithm — Leiden clustering, a state-of-the-art community detection algorithm
• Storage — Communities are stored in PostgreSQL with full-text search indexes
• Naming — LLM-based naming using entity context and relationship patterns
• Search — PostgreSQL full-text search with optional semantic search using embeddings
• Performance — Optimized for graphs with hundreds to thousands of nodes

What's Next

We're continuously improving the Graph Communities feature. Upcoming enhancements include:

• Hierarchical communities — Multi-level community structures for very large graphs
• Community comparison — Analyze relationships between different communities
• Custom community definitions — Manually define or adjust community boundaries when needed
• Export capabilities — Export community structures for external analysis

Try Graph Communities now