Knowledge graph builder

🕸️ Knowledge Graph Builder

Extract structured knowledge graphs from any text-based knowledge source.

Overview

Transform unstructured text into structured knowledge graph JSON format, enabling entity relationship mapping, timeline validation, fact verification, and graph visualization compatible with D3.js, Neo4j, and other graph tools.

When to Use This Skill

Use this skill when you need to:

• Extract entities and relationships from documents, reports, or notes
• Build knowledge graphs for visualization
• Validate timelines and chronological consistency
• Link claims to supporting quotes and evidence
• Prepare data for graph database import (Neo4j, etc.)
• Create interactive network visualizations

Input Sources

Works with any text-based knowledge source:

Structured Documents: Interview transcripts, meeting notes, research reports, case studies

Semi-Structured Data: Email threads, chat logs, wiki pages, knowledge base articles

Unstructured Text: Articles, essays, books, archived documents, legacy knowledge bases

Workflow

1. Read Source Material

Load your knowledge source(s) and note the source type and metadata.

2. Extract Entities

Identify key entities with confidence scoring:

• Person: Individuals (names, roles, relationships)
• Organization: Companies, institutions, groups
• Place: Locations, offices, geographic references
• Concept: Themes, topics, abstract ideas
• Claim: Factual assertions requiring verification

See references/entity-extraction.md for detailed patterns.

3. Identify Relationships

Map connections between entities using typed relationships (works_for, reports_to, collaborated_with, located_in, mentioned_in, etc.).

See references/relationship-types.md for complete taxonomy.

4. Extract Quotes & Evidence

Capture verbatim text with:

• Speaker/author attribution
• Temporal context (when)
• Spatial context (where)
• Sentiment analysis
• Confidence scoring

5. Link Claims to Evidence

Identify factual assertions, map supporting evidence (quote IDs), and calculate claim confidence based on evidence strength.

6. Validate & Quality Check

Run validation checks:

• ID uniqueness across all entities
• Referential integrity (all IDs resolve)
• Timeline consistency
• Evidence completeness (all relationships have quotes)
• No orphaned nodes

See references/validation-checklist.md for complete QA process.

7. Generate Outputs

Create visualization-ready files:

1. {name}-knowledge-graph.json - Structured graph data
2. {name}-graph-viewer.html - Interactive visualization (optional)
3. {name}-validation-report.md - Quality metrics

Output Format

JSON conforming to knowledge graph schema:

{
  "source": {
    "title": "Knowledge Source Title",
    "source_type": "document|transcript|report|article",
    "extracted_at_iso": "2026-01-31T00:00:00Z"
  },
  "entities": [/* people, orgs, places, concepts with aliases */],
  "themes": [/* recurring topics with keywords */],
  "quotes": [/* verbatim text with context and confidence */],
  "claims": [/* assertions with evidence */],
  "relationships": [/* typed connections with evidence */]
}

See examples/sample-output.json for complete anonymized example.

Output Location

Save files to project directory: /home/sasha/all-project-files/deployed-md-files/docs/{project-name}/

Confidence Scoring

All entities, relationships, quotes, and claims include confidence scores (0.0-1.0):

Scoring Factors:

1. Mention frequency - How often mentioned
2. Context clarity - Explicitly stated vs implied
3. Evidence strength - Supporting quotes available
4. Disambiguation - Unique vs ambiguous references
5. Source quality - Authoritative vs uncertain

Confidence Ranges:

• 0.90-1.00 - High confidence, multiple clear references
• 0.75-0.89 - Good confidence, explicit mention
• 0.60-0.74 - Moderate confidence, some ambiguity
• 0.40-0.59 - Low confidence, inferred or unclear
• <0.40 - Very uncertain, flag for review

Visualization Integration

The knowledge graph JSON integrates with the cool-charts skill for visualization:

Interactive Network Graph: Force-directed layout, color/size-coded, filterable

Timeline View: Chronological progression, duration bars, quote markers

See the cool-charts skill for complete visualization options.

Quality Standards

Schema Compliance: All required fields, correct data types, valid enum values

Evidence Requirements: Every relationship has evidence_quote_ids, all IDs reference existing objects, quotes are verbatim

Timeline Integrity: Chronologically consistent, no contradictions, temporal contexts align

Entity Normalization: Aliases captured, no duplicates, clear disambiguation

Graph Completeness: All entities have ≥1 relationship, no orphaned nodes, visualization-ready

Common Use Cases

Career/Professional Analysis: Extract people, companies, roles; map employment, reporting, collaboration; validate timeline

Research Documentation: Extract authors, concepts, institutions; map citations, collaboration, influence; link claims to evidence

Business Intelligence: Extract companies, products, markets; map partnerships, competition, acquisitions; track changes

Knowledge Base Migration: Extract structured data from legacy docs; preserve relationships and context; enable graph database import

Reference Documentation

Detailed extraction patterns: references/entity-extraction.md

Relationship taxonomy: references/relationship-types.md

Quality assurance: references/validation-checklist.md

JSON schema: references/schema.md

Troubleshooting

Ambiguous entity references: Use alias system, assign lower confidence, check context

Missing temporal context: Keep time_text as-is if relative, set time_iso to null, flag with certainty: "uncertain"

Too many potential entities: Focus on frequently mentioned (3+ references), prioritize entities with clear relationships

Orphaned entities: Only include entities with ≥1 relationship, document isolated mentions in notes

No evidence for relationship: Search for implicit evidence, mark as inferred if contextual, don't create if no evidence

Success Criteria

Functional: JSON validates, all IDs unique, all relationships have evidence, quotes are verbatim

Quality: Key entities identified with aliases, themes represented, timeline consistent, sentiment/certainty scored

Integration: Visualization-ready JSON, compatible with cool-charts skill, importable to graph databases

---

Skill Version: 2.0 (Generalized) Created: 2026-01-31 Based On: quote-to-knowledge-graph-extraction v1.0