ABSTRACT
The rapid expansion of online information poses significant challenges for individuals attempting to stay informed about current events, primarily due to the substantial time required to process extensive content. This challenge is particularly evident among professionals, students, and casual readers who require efficient ways to access critical information. Extractive text summarization, powered by Transformer-based models, provides an effective solution by compressing large volumes of text into concise summaries while retaining the core meaning of the original content. This research emphasizes political news, a domain known for its complexity and importance, utilizing the “BBC Political News Articles” dataset, which includes 417 articles and 2,085 pre-existing extractive summaries. The study assesses the performance of three Transformer-based Models—T5, BART, and PEGASUS—in news summarization tasks, employing ROUGE scores as the evaluation metric. The findings reveal that T5 delivers the highest performance with ROUGE-1, ROUGE-2, and ROUGE-L scores of 0.511, 0.281, and 0.466, respectively. BART follows with scores of 0.435, 0.201, and 0.390, while PEGASUS records lower scores of 0.154, 0.071, and 0.144. These results underscore T5 superior capability in generating precise and unbiased summaries for political news.
