knowledge-graph-construction_skill

This skill guides you in designing and building knowledge graphs from unstructured data, offering extraction pipelines, schema patterns, and data model

30

GitHub Stars

1

Bundled Files

2 months ago

Catalog Refreshed

4 months ago

First Indexed

Readme & install

Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.

Installation

Preview and clipboard use veilstrat where the catalogue uses aiagentskills.

npx veilstrat add skill lyndonkl/claude --skill knowledge-graph-construction

SKILL.md9.5 KB

Overview

This skill helps design and build knowledge graphs from unstructured or semi-structured data to support retrieval-augmented generation (RAG), reasoning, and analytics. It provides guidance on choosing a graph data model, defining schemas and ontologies, and configuring extraction pipelines. The goal is to produce traceable, high-quality graph facts that reduce hallucination and enable multi-hop queries.

How this skill works

I inspect your domain scope and input data to recommend an appropriate graph model (LPG, RDF/OWL, Hypergraph, Temporal). I produce a schema pattern (entity types, relation types, properties, constraints), an extraction pipeline (LLM-assisted NER, relation extraction, normalization), and a layered architecture with provenance. I also supply validation checks and a pragmatic output template you can use to implement and iterate your KG.

When to use it

Designing a KG for RAG or explainable retrieval
Choosing between LPG vs RDF/OWL for a new project
Building extraction pipelines from documents, logs, or literature
Modeling N-ary relations or time-evolving facts
Preparing a schema for ontology alignment or integration

Best practices

Scope narrowly first: list data sources, entity/relation priorities, and target queries
Prefer layered architecture: separate instance data, curated domain facts, and canonical ontology
Use hybrid extraction: LLM prompts + classical NER/dependency parsing, with a verification pass
Enforce provenance and confidence metadata on every triple or edge
Iterate schema after sampling extracted output and running consistency checks

Example use cases

Enterprise RAG: LPG with Neo4j + vector integration for fast, iterative deployment
Biomedical research: RDF/OWL for ontology alignment and formal reasoning with UMLS/SNOMED
Event modeling: Hypergraph or Temporal model for multi-participant events and time-series state
Legal/compliance: RDF/OWL with strong provenance chains and SPARQL queries
Scientific literature: layered approach — extract claims into Layer 1, curate into Layer 2, link to an ontology in Layer 3

FAQ

Choose an LPG (e.g., Neo4j) for rapid prototyping and easy vector search integration; switch to RDF/OWL later if you need formal reasoning or standards interoperability.

How do I ensure extraction quality at scale?

Combine few-shot LLM prompts with deterministic parsers, run a second-pass LLM for verification, sample for human review, and compute graph statistics (coverage, contradictions, orphan nodes).