Arabic Abstractive Summarization of Privacy Policies

Developed an AI system to automatically summarize complex Arabic privacy policies into concise, easy-to-read formats while preserving key legal information. The solution uses a fine-tuned AraT5 Transformer model, trained on a curated dataset of 9,300+ Arabic document–summary pairs, including policies aligned with Saudi PDPL standards. The system employs a multi-route summarization pipeline combining direct Arabic summarization, translation-based summarization, and semantic similarity scoring using SBERT to ensure summaries remain faithful to the original content. Key Features: Handles long, complex legal texts (up to 1,024 tokens) Produces high-quality, Modern Standard Arabic summaries Optimized for privacy policies but adaptable to other legal documents Incorporates Large Language Models (LLMs) for abstractive summarization Evaluated with ROUGE, BLEU, METEOR, and SARI metrics Impact: Makes Arabic privacy policies more accessible for users, improves comprehension, and supports compliance checks for organizations.