<![CDATA[Sirah Nabawiyah is a religious narrative text that contains numerous events, figures, locations, and chronological relationships. These characteristics mean that summarisation methods relying solely on word frequency or lexical similarity may overlook historically significant sentences. This study proposes Hybrid TextRank for the extractive summarisation of Indonesian-language Sirah Nabawiyah texts. The proposed method combines TF-IDF and BM25 similarities as graph edge weights and incorporates sentence position and historical entity density into the PageRank personalisation vector. The dataset comprises 312 subchapters from two parts of the Sirah Nabawiyah book. Quantitative evaluation was conducted on 157 subchapters from Part 1, which include human reference summaries, while 155 subchapters from Part 2 were used as a supporting corpus. Each method selected three sentences and was evaluated using ROUGE-1, ROUGE-2, ROUGE-L, and intra-summary similarity. Hybrid TextRank achieved F1 scores of 0.3353 for ROUGE-1, 0.1178 for ROUGE-2, and 0.2264 for ROUGE-L. Compared with standard TF-IDF-based TextRank, these results represent improvements of 6.28%, 12.73%, and 8.59%, respectively. Hybrid TextRank achieved the highest ROUGE-2 score, although Max-TFIDF slightly outperformed it on ROUGE-1 and ROUGE-L. These findings indicate that personalisation based on narrative structure and historical entities helps preserve important word pairs, while sentence selection still requires further optimisation to improve unigram coverage and overall sentence ordering.]]>
Copyrights © 2026
