<![CDATA[Sirah Nabawiyah is a religious narrative text containing numerous events, figures, locations, and chronological relationships. These characteristics may limit summarization methods that rely solely on word frequency or lexical similarity. This study proposes Hybrid TextRank for extractive summarization of Indonesian Sirah Nabawiyah texts. The method combines TF-IDF and BM25 similarities as graph edge weights and incorporates sentence position and historical-entity weighting into a personalized PageRank mechanism. The dataset comprises 312 subchapters from two parts of Ar-Raheeq Al-Makhtum. Quantitative evaluation was conducted on 157 subchapters from Part 1 with human reference summaries, while 155 subchapters from Part 2 were used as a supporting corpus. All methods generated three-sentence summaries and were evaluated using ROUGE-1, ROUGE-2, ROUGE-L, and intra-summary similarity. Using an empirically selected damping factor of 0.70, Hybrid TextRank achieved F1 scores of 0.3401, 0.1216, and 0.2340 for ROUGE-1, ROUGE-2, and ROUGE-L, respectively. Hybrid TextRank achieved the highest ROUGE-2 and ROUGE-L scores, whereas Max-TFIDF obtained the highest ROUGE-1 score. Statistical testing confirmed these metric-specific differences, while ablation analysis showed that the complete Hybrid TextRank configuration outperformed its partial component combinations. These findings indicate that combining hybrid lexical similarity with domain-aware personalization improves phrase-level and sequence-level similarity in extractive summarization of Sirah Nabawiyah texts.]]>
Copyrights © 2026
