Network Analysis

January 12, 2026 · View on GitHub

Interactive relationship graphs that visualize connections between entities using vis-network with physics-based layouts.

What It Does

IntellyWeave's network analysis system transforms entity relationships into interactive graph visualizations. It combines:

Relationship mapping - AI-powered identification of entity connections from document co-occurrences
Graph theory analysis - Identification of central nodes, clusters, and patterns
Interactive visualization - Physics-based graph layouts with drag-and-drop interaction

flowchart LR
    A[Extracted Entities] --> B[Relationship Mapping]
    B --> C[Network Analysis]
    C --> D[Graph Construction]
    D --> E[vis-network Rendering]

    style B fill:#6366f1,color:#fff
    style C fill:#6366f1,color:#fff
    style E fill:#6366f1,color:#fff

Use When

You need to visualize relationships between people, organizations, and places
You're investigating hidden connections in document collections
You want to identify central figures or organizations
You need to understand network structures and clusters
You're building intelligence link analysis

Prerequisites

IntellyWeave backend running
Documents with entities extracted
At least one LLM provider configured

How It Works

Network Analysis Pipeline

sequenceDiagram
    participant Docs as Documents
    participant Extractor as Entity Extractor
    participant Mapper as Relationship Mapper
    participant Network as Network Agent
    participant Frontend as Graph Component

    Docs->>Extractor: Extract entities
    Extractor->>Mapper: Entities with metadata
    Mapper->>Mapper: Find co-occurrences
    Mapper->>Network: Relationship map
    Network->>Network: Graph theory analysis
    Network-->>Frontend: Network findings
    Frontend->>Frontend: Render vis-network graph

Relationship Detection

The Relationship Mapper agent identifies connections through:

Document co-occurrence - Entities mentioned together in same chunks
Contextual analysis - LLM analysis of relationship nature
Strength scoring - Frequency and proximity of mentions

Network Analysis

The Network Agent performs graph theory analysis:

Centrality metrics - Identifies most connected nodes
Cluster detection - Groups of related entities
Pattern recognition - Recurring relationship structures

Visual Presentation

Network Pattern Analysis

The Network Agent displays text-based network intelligence findings with pattern descriptions and confidence scores.

Network Graph Visualization

Network Chart

Interactive vis-network graph showing entity relationships with physics-based layout.

Accessing Network View

Network View Callout

Access the network visualization from the sidebar to explore entity relationships.

Data Structures

NetworkPayload (Frontend)

type NetworkPayload = {
  title: string;           // Graph title
  description: string;     // Graph description
  nodes: NetworkNode[];    // Array of nodes
  edges: NetworkEdge[];    // Array of edges
  layout?: "hierarchical" | "random" | "circle" | "grid" | "force";
  layout_direction?: string;
  layout_sort?: string;
};

type NetworkNode = {
  id: string;              // Unique node identifier
  label: string;           // Display label
  type?: string;           // Entity type (person, org, location)
  color?: string;          // Node color
  size?: number;           // Node size
  group?: string;          // Grouping for clustering
  tooltip?: string;        // Hover text
  meta?: Record<string, unknown>;
};

type NetworkEdge = {
  id?: string;             // Edge identifier
  from_node: string;       // Source node ID
  to_node: string;         // Target node ID
  label?: string;          // Edge label
  strength?: number;       // Connection strength 0-1
  weight?: number;         // Visual weight
  directed?: boolean;      // Arrow direction
  tooltip?: string;        // Hover text
};

Network Finding (Backend)

{
    "name": "Vatican-CIC Intelligence Network",
    "type": "Organizational Network",
    "description": "Primary institutional structure facilitating escape operations",
    "assessment": "Central to rat line coordination with multiple connection points",
    "confidence": 0.92,
    "reasoning": "Document co-occurrences show repeated Vatican-CIC coordination"
}

Architecture

Backend Structure

backend/elysia/tools/intelligence/
├── network_agent.py           # NetworkAgent - graph theory analysis
├── mapper_agent.py            # MapperAgent - relationship detection
└── objects.py                 # IntelligenceMessage types

Frontend Structure

frontend/app/components/chat/displays/
└── ChartTable/
    └── NetworkDisplay.tsx     # vis-network rendering

Key Components

Component	File	Purpose
NetworkAgent	`tools/intelligence/network_agent.py`	Text-based network analysis
MapperAgent	`tools/intelligence/mapper_agent.py`	Relationship detection
NetworkDisplay	`ChartTable/NetworkDisplay.tsx`	vis-network rendering

Analysis Capabilities

Network Agent Output

The NetworkAgent provides text-based intelligence findings, not raw graph data:

class NetworkAnalysisSignature(dspy.Signature):
    """Analyze intelligence relationship networks..."""

    network_findings: List[Dict] = dspy.OutputField(
        desc="""List of network intelligence findings:
        - name: Network pattern name
        - type: Category (Network Structure, Person-to-Person, etc.)
        - description: What the network represents
        - assessment: Why it's significant
        - confidence: 0.0-1.0
        - reasoning: Evidence supporting this finding"""
    )

Finding Types

Type	Description
Network Structure	Overall network topology
Person-to-Person	Direct interpersonal connections
Organizational Network	Institutional relationships
Adversary Network	Opposing or competing networks

vis-network Features

Physics-Based Layout

The graph uses ForceAtlas2 physics simulation:

Nodes repel each other naturally
Edges act as springs
Related nodes cluster together
Interactive drag-and-drop repositioning

Layout Options

Layout	Description
`force`	Physics-based (default)
`hierarchical`	Tree-like structure
`circle`	Nodes arranged in circle
`grid`	Regular grid pattern
`random`	Random positioning

Node Styling

Nodes can be styled by entity type:

// Color by entity type
const nodeColors = {
  person: "#ef4444",      // Red
  organization: "#3b82f6", // Blue
  location: "#10b981",     // Green
};

Fullscreen Mode

Click the fullscreen button to expand the network graph for detailed analysis.

Integration with Intelligence Analysis

Network Analysis is Phase 4 of the 6-phase Intelligence Orchestrator:

flowchart TB
    subgraph Intelligence Orchestrator
        P1[1. Entity Extractor]
        P2[2. Relationship Mapper]
        P3[3. Geospatial Analyst]
        P4[4. Network Analyst]
        P5[5. Pattern Detector]
        P6[6. Synthesizer]
    end

    P1 --> P2
    P2 --> P3
    P3 --> P4
    P4 --> P5
    P5 --> P6

    style P4 fill:#6366f1,color:#fff

The Network Agent receives: