Arifov's picture
Upload README.md with huggingface_hub
bd374b7 verified
---
pretty_name: ECHR Guide Paragraph Diffs
language:
- en
tags:
- legal
- echr
- temporal
- paragraphs
- dataset-creation
size_categories:
- 10K<n<100K
---
# Dataset Card for ECHR Guide Paragraph Diffs
## Dataset Summary
This dataset contains paragraph-level and section-level diffs between consecutive versions of European Court of Human Rights case law guide PDFs.
Each JSON file covers one guide version transition and records paragraph-by-paragraph changes (additions, deletions, edits, moves), section structural events (additions, deletions, renames), and cited-case list diffs.
It is the primary source corpus used to build `lexgenie/echr-case-linked-guide-diffs` — the modeling-ready citation-change layer.
## Dataset Structure
Files are organized by guide directory (39 guide directories, 145 JSON files total):
```
<guide_id>/
diff_<date_a>__<date_b>.json
...
```
Each JSON file has the following top-level keys:
- `version_a`, `version_b`: ISO date strings for the two guide snapshots
- `delta_days`: number of days between versions
- `guide_title`: human-readable guide name
- `summary`: aggregate counts by change type
- `section_events`: list of section-level structural events
- `paragraph_changes`: list of paragraph-level change records
- `cited_cases_diff`: `{added: [...], removed: [...]}` citation list diffs
### Paragraph change record fields
| Field | Description |
|---|---|
| `change_type` | One of: `unchanged`, `minor_edit`, `citation_added`, `citation_updated`, `citation_removed`, `reformulation`, `paragraph_added`, `paragraph_deleted`, `section_moved`, `section_moved_modified` |
| `section_path` | Hierarchical section identifier |
| `section_title` | Section heading text |
| `section_level` | Nesting depth |
| `para_num_a`, `para_num_b` | Paragraph index in version A and B |
| `similarity` | Text similarity score (0–1) |
| `text_a`, `text_b` | Paragraph text before and after |
| `citations_added`, `citations_removed` | Per-paragraph citation deltas |
| `moved_from`, `moved_to` | Section move provenance (when applicable) |
## Dataset Statistics
- Guide directories: `39`
- Diff files: `145`
- Total paragraph changes (all types): `39148`
- Of which minor edits: `6342`
- Of which paragraph additions: `1877`
- Of which paragraph deletions: `1418`
- Of which citation additions: `2147`
- Section events: `689` (238 added, 237 deleted, 192 renamed, 22 title-changed)
- Cited cases added across all transitions: `2987`
- Cited cases removed across all transitions: `1084`
- Total size: ~83 MB
## Intended Use
This dataset is intended for:
- temporal guide-body analysis across ECHR case law guides
- paragraph-level change classification
- section localization for case-linked guide updates
- training and evaluation of guide-update generation models
## Provenance
Produced by Anas Belfathi from Wayback Machine PDF snapshots of ECHR-KS guides (2022–2025). Paragraph alignment and diff computation use a custom PDF-parsing and text-alignment pipeline.
The guide snapshot archive is not publicly distributed due to size and licensing considerations; the Wayback Machine URLs are documented in `wayback/*/snapshot_metadata.jsonl` in the `lexgenie/lexgenie` repository.
See `lexgenie/echr-guide-citation-diffs` for the underlying PDF snapshot metadata and citation-list diffs, and `lexgenie/echr-case-linked-guide-diffs` for the modeling-ready downstream layer.