CSVW-EO Overview¶
CSVW-EO extends the W3C CSV on the Web (CSVW) standard with privacy-safe metadata for:
- Differential Privacy (DP)
- Dummy data generation
- Structural dataset modeling
- Public partition definitions
- Safe schema publication
CSVW-EO allows organizations to publish assumptions and guarantees about datasets without exposing sensitive underlying records.
These assumptions may include:
- dataset schema
- nullable proportions
- public categorical domains
- grouping partitions
- contribution bounds for DP
- logical dependencies between columns
Warning
Some assumptions may themselves leak sensitive information. Metadata must always be manually reviewed before publication.
Main Concepts¶
CSVW-EO extends CSVW with:
| Concept | Purpose |
|---|---|
| Structural modeling | Describe possible datasets |
| Dummy modeling | Generate realistic fake datasets |
| DP contribution bounds | Calibrate differential privacy |
| Public partitions | Define safe grouping assumptions |
| Validation | Ensure metadata consistency |
Related Components¶
| File | Purpose |
|---|---|
csvw-eo-vocab.ttl |
RDF vocabulary |
csvw-eo-context.jsonld |
JSON-LD context |
csvw-eo-constraints.ttl |
SHACL validation |
csvw-eo-library |
Python tooling |