Automatic Summarization for Arabic Scientific Articles Using Combined Approaches

Summarizing Arabic text from articles is a crucial step in Natural Language Processing (NLP) and remains an ongoing challenge. The presence of graphical elements such as graphs and figures further complicates the development of an effective and precise automatic text summarization algorithm. This paper presents a simple and robust summarization algorithm for Arabic scientific articles. Our approach consists of two stages: preprocessing and text summarization. It successfully extracts article summaries, even when the pages have a complex structure. Our method combines projection profiles, a feature-based OCR system, a hybrid approach that utilizes BART sequence-to-sequence model, cosine similarity, graph construction, the PageRank algorithm, and sentence weight score assignment. The presented method demonstrates efficient performance, mainly when dealing with non-uniform articles where the page structure varies across different pages. We conducted experiments on a collected dataset to evaluate our approach. The results of these experiments indicate that our proposed method is highly efficient compared to existing methods.