Graph Visualization and Search

Interactive exploration of knowledge graphs with filtering and search

sift-kg provides two ways to explore your knowledge graph: interactive visualization with sift view and command-line search with sift search.

Interactive Visualization

sift view

Generates an interactive HTML visualization and opens it in your browser.

Command Options

-o, --output · path

Output directory containing graph_data.json

--to · path

Custom path for the HTML file (defaults to output/graph.html)

--no-open · boolean

Generate HTML but don’t open browser

-v, --verbose · boolean

Verbose logging

Filtering Options

--top · integer

Show only the top N entities by connection count (and their direct neighbors)

--min-confidence · float

Hide entities and relations below this confidence threshold (0.0-1.0)

--source-doc · string

Show only entities from a specific document

--neighborhood · string

Center view on a specific entity and show N-hop neighborhood

--depth · integer · default: "1"

Number of hops for neighborhood view (use with --neighborhood)

--community · string

Focus on a specific community cluster (e.g., “Community 1”)

Visualization Examples

Basic View

# Full graph
sift view

# Custom output location
sift view --to ./reports/knowledge-graph.html

# Generate without opening
sift view --no-open

Filtered Views

Top Entities

# Show 50 most connected entities
sift view --top 50

# Top 20 entities, high confidence only
sift view --top 20 --min-confidence 0.8

Interactive Features

The visualization includes a rich sidebar with:

Search

Type to find entities by name (case-insensitive, fuzzy matching).

Entity Type Toggles

Show/hide entity types:

• Click checkbox to toggle visibility
• Click color picker to change type color
• All changes update graph in real-time

Relation Type Toggles

Control which relations are displayed:

• Show/hide by relation type
• Relation count shown for each type
• MENTIONED_IN hidden by default (metadata)

Community Filters

Toggle visibility by community cluster:

• Communities auto-detected using Louvain algorithm
• Border colors indicate community membership
• Filter to single community for focused analysis

Detail Panel

Click any node to see:

• Entity name and type
• All attributes
• Aliases
• Source documents
• Generated description (if sift narrate was run)
• All incoming/outgoing relations

Click any edge to see:

• Relation type
• Confidence score
• Evidence text
• Support count (how many times extracted)
• Support documents

Dynamic Filters

Degree Filter: Hide low-connectivity entities

Min connections: [slider] 0 ━━━●━━━ 10

Confidence Filter: Show only high-quality extractions

Min confidence: [slider] 0 ━━━━━●━ 1.0

Source Document Count: Entities mentioned in multiple docs

Min sources: [slider] 1 ━●━━━━━ 5

Source Document Selector: Dropdown to filter by specific document

Physics Controls

Adjust graph layout:

• Pause/resume physics simulation
• Freeze positions after stabilization
• Reset zoom and positions

Visualization Colors

Entity Colors (by Type)

Predefined semantic colors for common types:

• PERSON: Blue
• ORGANIZATION: Green
• LOCATION: Purple
• EVENT: Orange
• DOCUMENT: Gray

Other types auto-assigned from palette with maximum hue separation.

Relation Colors (by Type)

Semantic colors for common relations:

• EXTENDS: Purple (inheritance)
• SUPPORTS: Green (evidence for)
• CONTRADICTS: Red (evidence against)
• USES_METHOD: Blue (methodology)
• MENTIONED_IN: Dark gray (low emphasis)

Other types auto-assigned from edge color palette.

Community Colors

Community border colors use golden angle spacing for maximum visual separation.

Search Command

sift search "John Smith"

Command-line search for entities by name or alias.

Command Options

query · string · required

Search term (matches entity names and aliases, case-insensitive)

-r, --relations · boolean

Show connected entities and their relations

-d, --description · boolean

Show entity description (requires sift narrate to have been run)

-t, --type · string

Filter results by entity type (e.g., PERSON, ORGANIZATION)

-o, --output · path

Output directory containing graph_data.json

-v, --verbose · boolean

Show all relations (default limits to 10)

Search Examples

Basic Search

# Find entities matching "Smith"
sift search Smith

# Case-insensitive, searches names and aliases
sift search "acme corp"

Example Output

$ sift search "John Smith" --relations

3 results

  PERSON: John Smith
    aka: J. Smith, Dr. Smith
    Connections: 15
    Sources: document1, document2, document5
    → WORKS_FOR → Acme Corporation
    → FOUNDED → SmithCo Industries
    → COLLABORATED_WITH → Jane Doe
    ← SUPERVISED → Bob Johnson
    ← HIRED → Acme Corporation
    ... 10 more (use --verbose to show all)

  PERSON: Bradley Smith
    Connections: 3
    Sources: document7
    → ATTENDED → Stanford University

Use Cases

Investigative Research

Scenario: Finding connections between people and organizations

# Find all mentions of a person
sift search "John Doe" --relations --description

# View their network
sift view --neighborhood "John Doe" --depth 2

# Focus on high-confidence connections only
sift view --neighborhood "John Doe" --min-confidence 0.8

Document-Specific Analysis

Scenario: Analyzing what was extracted from one document

# View entities from specific report
sift view --source-doc "annual-report-2023.pdf"

# Search within that document
sift search "revenue" --type CONCEPT

Community Detection

Scenario: Finding clusters of related entities

# Generate view with communities
sift view

# In browser: check community colors on entity borders
# Export communities.json for external analysis

# Focus on specific community
sift view --community "Community 3"

Quality Assurance

Scenario: Checking extraction quality

# View low-confidence entities (may need review)
sift view --min-confidence 0.5

# Find isolated entities (no relations)
sift view --top 200  # Then filter by degree = 0 in sidebar

# Check entities with few source documents
# In sidebar: set "Min source docs" to 1

Performance Tips

1. Filter Large Graphs

   For graphs with >1000 entities, use filters to reduce visual complexity:

   sift view --top 100
   sift view --min-confidence 0.7
   
2. Use Neighborhoods

   Focus on specific areas instead of viewing entire graph:

   sift view --neighborhood "key-entity" --depth 2
   
3. Hide Metadata Relations

   MENTIONED_IN edges clutter large graphs. They’re hidden by default, but can be shown via the sidebar if needed.

4. Freeze Physics

   In the visualization, pause physics after initial layout stabilizes to improve browser performance.

Saving and Sharing

The generated HTML is fully self-contained:

# Generate static HTML
sift view --to ./report.html --no-open

# Share via email, upload to web server, or include in reports
# No sift-kg installation needed to view

All JavaScript, CSS, and graph data embedded in a single file.

Troubleshooting

Blank or empty visualization

Cause: All entities filtered out

Solution: Check filters and graph size:

sift view --verbose  # Shows entity/relation counts after filtering

Relax filters:

sift view --min-confidence 0.0 --top 1000

“Graph not found”

Run sift build first to create graph_data.json.

Slow or laggy browser performance

For large graphs (>500 entities):

• Use filtering: --top 200 --min-confidence 0.7
• Pause physics after stabilization
• Use neighborhood view instead of full graph
• Close detail panel when not needed

Search returns no results

• Check spelling (search is case-insensitive but must match name/alias)
• Try partial terms: sift search Smith instead of sift search "John Smith"
• Check entity type filter: --type may be excluding results
• Verify graph contains entities: sift search "" --verbose

Communities not showing

Communities auto-detected on first sift view. If missing:

# Rebuild communities
sift narrate --communities-only  # Fast, ~$0.01

# Or regenerate full view
sift view

Next Steps

Export Data

Export graphs for external analysis tools

Generate Narratives

Create AI-generated summaries of your knowledge graph