Creation of an automatic summary tool for arabic texts

Abstract

This thesis presents an advanced Arabic text summarization system built by fine tuning a T5-base transformer model on the AGS-Corpus, a curated Arabic summarization dataset. The model effectively handles the language’s rich morphology, diverse dialects, and complex syntax. Evaluation using BERTScore and ROUGE shows that the system generates high-quality summaries closely aligned with human references in content, fluency, and semantic similarity. To improve accessibility, the system is deployed as a user-friendly web application using Streamlit, enabling real-time summarization for non-technical users. Future work includes expanding the dataset to include more dialects and exploring alternative architectures like BART or GPT to further enhance performance, especially on complex or lengthy texts