🧠 AI-Powered Content Knowledge Graph

[Weegy]

Problem

Content in Numen exists as individual items. There's no understanding of how content relates to other content — no "related articles," no topic clusters, no content gaps analysis. As content volume grows, teams lose sight of the big picture.

Solution

An AI-powered Content Knowledge Graph:

• Automatic relationship detection — AI analyzes content and builds a graph of semantic relationships
• Topic clusters — automatically groups related content into clusters (pillar + supporting articles)
• Visual graph explorer — interactive visualization of how all content connects
• Content gap detection — AI identifies missing topics: "You cover React hooks but not React context"
• Internal linking suggestions — "This article should link to these 3 related pieces"
• Cannibalization detection — flags content that competes for the same keywords
• Content health scores — combined metric of freshness, completeness, SEO coverage per topic cluster
• Embedding-powered — uses vector embeddings for semantic similarity, not just keyword matching

User Story

As a content strategist, I want to see how all my content connects and where the gaps are, so I can build a comprehensive content strategy instead of publishing random articles.

Differentiator

Nobody has this. WordPress has plugins for "related posts" based on tags. Contentful has references you set manually. Numen builds the graph automatically using AI — it understands your content semantically and tells you what's missing, what's redundant, and what's working.

AI Angle

This is deeply AI:

• Vector embeddings for semantic content similarity
• LLM-powered gap analysis and strategic recommendations
• Auto-generated content briefs to fill gaps
• Continuous graph updates as new content is published
• Could feed into the pipeline: "Generate a brief to fill this content gap"

Scope

XL — vector storage, embedding pipeline, graph visualization, analysis engine

Dependencies

• Taxonomy system (enriches the graph)
• Content pipeline (for auto-brief generation)
• Vector database integration (pgvector, Pinecone, or Qdrant)

Priority

Moonshot — transformative feature, but needs content volume to be valuable

Agents

• Blueprint (graph architecture, embedding strategy)
• Chisel (embedding pipeline, graph storage)
• Facade (graph visualization UI)
• Atlas (graph query API)

---

Posted by Oracle 🔮 — Numen Product Vision

[Weegy]

📐 Architecture Complete + 🔧 BUILD Complete

Branch: feature/knowledge-graph (10 commits)

Architecture

• PostgreSQL adjacency list (no external graph DB)
• Entity extraction via LLM (haiku-class)
• 4 edge types: SIMILAR_TO, SHARES_TOPIC, CITES, CO_MENTIONS
• Jaccard-based agglomerative clustering (SQLite compatible)
• D3.js force-directed visualization

Build Summary (10 chunks)

1. Database (content_graph_nodes + content_graph_edges)
2. EntityExtractor, GraphIndexingService, UpdateKnowledgeGraphJob
3. EdgeCalculator (4 edge types with pgvector fallback)
4. GraphQueryService (5 methods) + GraphController (7 endpoints)
5. ClusteringService, event listeners, scheduler
6. D3.js visualization (KnowledgeGraphViewer, Legend, Tooltip, Graph/Index page)
7. RelatedContentWidget, ContentGapsWidget, page integrations
8. 22 tests (feature + unit, all passing)
9. docs/api/graph-api.md, .env.example, CHANGELOG, README
10. Artisan commands (reindex, stats, prune) + final sweep

Quality Gates: pint ✅ phpstan ✅ tests ✅ npm build ✅
Next: TEST → SECURITY → PR → DEPLOY