SummARai: Arabic Book Summarization Platform

🌟 Introduction

SummARai (Summarizing Arabic with AI) is an innovative web-based platform designed to automatically generate high-quality summaries for Arabic books and documents using advanced AI technologies. As a graduation project from Cairo University's Faculty of Computers and Artificial Intelligence, SummARai addresses the critical gap in Arabic Natural Language Processing (NLP) tools by providing a specialized solution for Arabic content summarization.

🚀 Key Features

• Hybrid Summarization Pipeline: Combines extractive(Graph-Based) and abstractive(AraT5) techniques for optimal results
• Arabic Language Specialization: Specifically designed for Arabic's complex morphology and rich syntax
• User-Friendly Interface: Intuitive web platform with Arabic UI
• Advanced AI Models: Leverages AraT5 transformer and LLaMA 4 for superior summarization
• Comprehensive Library: Searchable database of book summaries
• Reading Progress Tracking: Personal statistics and reading history
• OCR Support: Processes both digital and scanned PDFs

🔧 Technology Stack

Frontend

• React.js
• Tailwind CSS
• Axios for API communication

Backend

• Java Spring Boot
• MySQL Database
• Redis Caching
• JWT Authentication

Machine Learning

• Python FastAPI
• AraT5v2-base-1024 (fine-tuned)
• LLaMA-4-Scout-17B-16E-Instruct
• Mistral AI OCR
• Sentence Transformers for semantic chunking

Infrastructure

• AWS Cloud Services
• Theta Labs GPU Resources (NVIDIA H100/A100)

📊 Performance Highlights

Our hybrid approach outperforms traditional methods:

Metric	Our Model	Baseline (mT5)
BERTScore Precision	77.55%	70.76%
BERTScore Recall	66.54%	63.98%
BERTScore F1	71.73%	61.72%

🛠️ Installation & Setup

Prerequisites

• Node.js (v16+)
• Java JDK (v17+)
• Python (v3.9+)
• MySQL (v8.0+)
• Redis

Frontend Setup

cd Frontend
npm install
npm run dev

Backend Setup

cd Backend
./mvnw spring-boot:run

ML Service Setup

cd Model/ModelPreprocessig
pip install -r requirements.txt
uvicorn main:app --reload

📚 Dataset

We created SummARai v1.0 - a comprehensive dataset of 4,328 Arabic text chunks with aligned summaries, covering various genres. Available at:
SummARai Dataset GitHub

🤝 Team

• Fatma Elzahraa Ashraf Samy
• Rowan Madeeh Mohamed
• Abdelrahman Mohamed Ezzat
• Abdelrahman Gomaa
• Momen Mostafa Mabrouk

Supervised by:
Dr. Mohammed ElRamly
TA. Amany Hesham

🌐 Live Demo

Explore SummARai: Demo Link

---

SummARai - Bridging the gap between Arabic literature and AI-powered accessibility.