GraphFaker¶
Synthetic graph data that behaves like the real thing. Generate realistic graph datasets from a schema, or pick a ready-made domain such as a bank with laundering patterns, and get entities, relationships and events whose structure, attributes and timing agree, with the ground truth included.
Contents:
- graphfaker
- Installation
- Quick Start Guide
- Usage
- How generation works
- How the fraud graph is generated
- Ways to generate synthetic graph data
- Adding a domain
- GraphFaker tour: schemas, a synthetic bank, and the ground truth that comes with it
- Measuring entity duplication in an LLM-built knowledge graph
- What this measures
- What it does not measure
- Requirements
- Step 1 — Build a corpus with a known answer key
- Step 2 — The control
- Step 3 — Run Cognee
- Step 4 — Measure
- Step 5 — Repair the graph, then re-measure
- Step 6 — Load the repaired graph into a database
- Step 7 — Record the result
- Caveats to carry into any write-up
- Contributing
- Credits
- History