ArabiCCR

A pipeline for scraping, preprocessing, and anonymizing Arabic judicial rulings from the Saudi Ministry of Justice.

---

Step-by-step instructions:

1. Download the dataset from [https://data.mendeley.com/drafts/np538c95yy]
2. Install dependencies (requirements.txt)
3. Run preprocessing/redaction scripts as described in the configuration section

---

Pipeline

MOJScraper.py  →  ColumnSplit.py  →  NameRedaction.py

Step	Script	Input	Output
1	MOJScraper.py	MOJ website	output1.csv, output2.csv, …
2	ColumnSplit.py	.xlsx from step 1	.xlsx with EVENTS, REASONING, RULING columns
3	NameRedaction.py	.xlsx from step 2	.xlsx with person names replaced by [Person Name]

---

Setup

Requirements: Python 3.9+, Google Chrome, ChromeDriver matching your Chrome version, and Stanza Arabic models.

pip install -r requirements.txt

Download Stanza Arabic models (one-time):

import stanza
stanza.download("ar")

---

Configuration

All settings are in config.py. Edit it before running any script:

• Set CHROME_DRIVER_PATH to your local ChromeDriver executable.
• Set STANZA_MODEL_DIR to where Stanza models are saved.
• Set input/output file paths for each step.
• Adjust COURT_TYPE to target a different court category.

---

Usage

Run each script in order:

python MOJScraper.py
python ColumnSplit.py
python NameRedaction.py

MOJScraper.py saves a checkpoint.json and resumes automatically if interrupted.

---

Notes

• Scraper output is split into chunks of 1,000 judgments per CSV (configurable via CHUNK_SIZE).
• Redaction uses two passes: role pattern matching followed by Stanza NER, then a global sweep for any names.
• Only columns specified in TARGET_COLUMN_LETTERS are redacted.