<p>The exponential growth of Arabic digital content has intensified the challenge of information overload, highlighting the need for effective Automatic Text Summarization (ATS) techniques. Existing graph-based approaches, such as TextRank and LexRank, demonstrate promising results but face notable limitations in Arabic, including shallow semantic modeling, neglect of sentence relevance, structural bias, and slow convergence. To address these challenges, this work introduces a hybrid graph–feature-based framework with four key innovations: (i) a refined graph construction strategy that integrates semantic features and sentence relevance into edge weights, (ii) a novel ranking algorithm, ST-Rank, that incorporates semantic information, achieves faster convergence through informed initialization, and leverages two new normalization strategies, (iii) a hybrid fusion mechanism that combines graph-based scores with semantic and statistical sentence-level attributes, and (iv) the design of novel feature sets to enhance sentence salience estimation. Extensive experiments on the Essex Arabic Summaries Corpus (EASC) demonstrate that the proposed approach surpasses current state-of-the-art summarization methods, achieving ROUGE-1 and ROUGE-2 F1-scores of 0.705 and 0.651, respectively, with significant gains over baseline and related techniques. Furthermore, ST-Rank exhibits markedly improved computational efficiency, converging up to five times faster than conventional graph-based methods. These results confirm the effectiveness and robustness of the proposed framework in advancing Arabic text summarization.</p>