← Back to projects
CompletedOctober 2025
Natural Language Processing for Algerian Legal Document Analysis and Automated Regulatory Compliance Checking
The Algerian legal corpus (Official Journal, codes, decrees) is voluminous, multilingual (Arabic/French), and poorly digitally structured. This project develops a complete NLP pipeline for legal entity extraction, cross-reference resolution between articles, and automatic compliance verification of administrative documents against current regulatory texts.
Team
AM
Abdelhamid MASSERITI
Research Director
FB
Farid BENHAMMADI
Researcher
KT
Karim TOUATI
Researcher
Methodology
- Build a parallel Arabic-French corpus of 12,000 articles from the Official Journal of the Algerian Republic (JORADP) with manual annotation of legal entities (law, decree, article, date, institution).
- Fine-tune CamemBERT (French) and AraBERT (Arabic) models for legal Named Entity Recognition (Legal NER) with cross-evaluation.
- Develop a cross-reference resolution module based on rules and a legal knowledge graph (links between articles, repeals, amendments).
- Prototype a compliance checker: given an administrative document (permit, contract), the system identifies clauses potentially non-compliant with current legislation.
Evaluation metrics
- F1-score on legal entity recognition — target > 88% (French) and > 82% (Arabic).
- Cross-reference resolution accuracy — target > 90% on a test set of 500 article-reference pairs.
- Inter-annotator agreement (Cohen's kappa) on the corpus — target > 0.85.
Budget
$500 in compute credits for language model fine-tuning, $400 for corpus annotation (law students from the University of Oran).
Expected deliverables
- Annotated JORADP-NER corpus published under open license.
- Self-hostable compliance verification REST API.
- Paper published in the Journal of Artificial Intelligence Research (JAIR).