Subhan Subhan, Novita Ayu Indirawati, Yahya Nur Ifriza
The rapid evolution of technology has fundamentally reshaped the landscape of digital news dissemination. However, the continuous surge in available data has precipitated a state of information overload, forcing readers to navigate vast quantities of content - a process that often undermines overall cognitive efficiency. Consequently, automated text summarization has emerged as a pivotal solution for optimizing the consumption of information. This research utilizes the TextRank algorithm, integrated with FastText word embeddings and cosine similarity, to transform textual input into vector representations. This methodology facilitates the ranking of sentences based on their thematic relevance, ensuring that the generated summaries remain both precise and concise. To validate the system's efficacy, the ROUGE metric was employed to conduct a comparative analysis between the system-generated output and human reference summaries. Empirical results demonstrated high performance, with F-scores of 0.92, 0.87, and 0.92 for ROUGE-1, ROUGE-2, and ROUGE-L, respectively. Optimal outcomes were achieved by configuring the system to a maximum of ten sentences and applying a similarity threshold of 0.4 to govern the connectivity within the sentence network. This specific calibration allowed for a sophisticated balance between brevity and comprehensiveness, yielding high-quality summaries that closely mirror the structural and substantive integrity of the reference texts. © 2026 IEEE.
Universitas Negeri Semarang, Computer Science Department, Semarang, Indonesia
