Arabic Headline Summarization Using Arat5

Headlines are regarded as a brief summary of the text. The necessity to manage a huge number of papers has raised the need to automatically generate headings. Hence, rather than reading every document, you may simply look at the headline to see which ones contain significant and pertinent information. Deep learning techniques have advanced this field of study, particularly as pre-trained models become more prevalent that have produced the finest outcomes. In this study, we used a T5 text-to-text transformer, which produced outcomes that were deemed ideal for a variety of tasks involving natural language processing. We used the AHS dataset containing 300,000 articles and their headlines. ROUGE-I, ROUGE-2, ROUGE-L, and BLEU model evaluation results were 0.53, 0.3, 0.36, and 0.48, respectively. The results achieved by the AraT5 model were better than those achieved by the Seq2seq model.