Elamur Research⌕
☰
home / papers / 2026 / Linking Hadith Narrator Identities Across Heterogeneous Arabic Biographical Databases: A Multi-Signal Entity Resolution Pipeline

Linking Hadith Narrator Identities Across Heterogeneous Arabic Biographical Databases: A Multi-Signal Entity Resolution Pipeline

Summary

Hadith narrators are recorded in several independently built digital resources with no shared identifiers. The same person can appear under a kunyah, a long nasab chain, with or without diacritics, or in different grammatical cases. This paper builds an entity resolution pipeline that connects three such resources. Every name first passes through an Arabic normalization designed for historical personal names. In Phase 1, the 185,216 narrator name variants of Sanadset 650K are matched on names alone to the Hadithtransmitters (hawramani) database, because Sanadset has no biographical fields. In Phase 2, hawramani entries are matched to Muslimscholars with a weighted score that combines name similarity, closeness of death years and the polarity of reliability grades. Phase 1 links about half of the Sanadset names, and Phase 2 links nearly all of hawramani. Chaining the two phases gives Sanadset narrators indirect access to Muslimscholars metadata and produces an enriched directed transmission graph. The link sets and graph are openly released. They are the first sizable cross-source resource of this kind and could serve as training data for future learned linkers.

Key points

  • Sources: Sanadset 650K (650,986 records from 926 books, about 4.8M narrator–Hadith occurrences), hawramani (100,915 entries, collected by BFS crawling over teacher/student links) and Muslimscholars (25,247 profiles).
  • Phase 1 yields 94,628 links (51.1%), split into HIGH 39,938 (name score ≥ 0.90) and MEDIUM 54,690. Phase 2 yields 95,573 links (94.7% of hawramani), split into HIGH 18,245, MEDIUM 71,546 and LOW 5,782.
  • The resulting graph has 185,216 nodes and 814,093 edges. The author describes it as the largest published Hadith narrator network by node count. The released resources are the sanad_links and narrator_links CSVs plus the graph, on Zenodo.
  • 88,343 Sanadset names (47.7%) get no bigram candidates at all. The main causes are uneven nasab depth and kunyah/ism alternation.
  • MEDIUM dominates Phase 2 mostly because data are missing: in just 4.8% of matched pairs is a death year available for both records. There is no gold evaluation set, and the tier thresholds were set by inspection.

Interesting ideas

  • I1. An early normalizer used too wide a Unicode range and erased all Arabic letters, so every record matched every other one; a sudden jump to near-total matching is a symptom worth testing for. — For Mova (learn): Add a negative control to evaluation that flags degenerate outcomes, such as all mentions collapsing to one entity or every pair agreeing, as a likely pipeline bug rather than success.
  • I2. Candidate generation indexes each name by its consecutive token bigrams, which cuts the comparisons by three to four orders of magnitude before any expensive scoring runs. — For Mova (world): Linking text entities to Wikidata/WordNet entries could use a cheap token-bigram index to shortlist candidates, then apply explainable scoring only to that shortlist.
  • I3. Which signals are combined depends on what metadata each pair actually has, and the confidence tier openly reflects missing evidence rather than hiding it. — For Mova (world): Entity and state links in the world of a text could record which signals supported them, so an answer can explain whether a link is weak because of ambiguity or because evidence was missing.
  • I4. Two pairwise linkings are composed transitively, so a resource without metadata inherits facts from a third database it was never directly matched to. — For Mova (world): Mova could chain alignments (for example, text mention → WordNet → Wikidata) to enrich entities, while keeping the tier of each hop visible.
  • I5. As future work, the author proposes graph constraint propagation: a confidently matched narrator restricts the candidates for his teachers to those documented for him. — For Mova (frames): In coreference, a confidently resolved anchor could narrow the antecedent choices for related mentions (his teacher, her son) by using known relations in the frame.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2607.05424 · arXiv 2607.05424

@misc{wirahman2026linking,
  title = {{Linking Hadith Narrator Identities Across Heterogeneous Arabic Biographical Databases: A Multi-Signal Entity Resolution Pipeline}},
  author = {Wirahman, Taufiq},
  year = {2026},
  doi = {10.48550/arxiv.2607.05424},
  eprint = {2607.05424},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2607.05424}
}

References

20 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Tanvir Alam and Jens Schneider. Social network analysis of Hadith narrators from Sahih Bukhari, 2021. URL https://arxiv.org/abs/2102.02009. ↗
  2. Wissam Antoun, Fady Baly, and Hazem Hajj. AraBERT: Transformer-based model for Arabic language understanding, 2020. URL https://arxiv.org/abs/2003.00104. ↗ our card
  3. Peter Christen. Data Matching: Concepts and Techniques for Record Linkage, Entity Resolution, and Duplicate Detection. Springer, 2012. doi: 10.1007/978-3-642-31164-2. ↗
  4. Doaa S. Elzanfaly and Ahmed S. Hassan. Arabic named entity disambiguation using linked open data. In 2015 IEEE International Conference on Signal and Image Processing Applications (ICSIPA), 2016. doi: 10.1109/ICSIPA.2015.7412226. ↗
  5. Doaa S. Elzanfaly and Ahmed S. Hassan. Arabic knowledge graph construction: A close look in the present and into the future. Journal of King Saud University – Computer and Information Sciences, 2022. doi: 10.1016/j.jksuci.2022.03.006. ↗
  6. Ikram Hawramani. Hadith Transmitters: Biographical database of Hadith narrators, 2026. URL https://hadithtransmitters.hawramani.com. Accessed June 2026. ↗
  7. Amna Kamran, Badar Abro, and Amna Basharat. SemanticHadith: An ontology-driven knowledge graph for the Hadith corpus. Web Semantics: Science, Services and Agents on the World Wide Web, 78:100797, 2023. doi: 10.1016/j.websem.2023.100797. ↗
  8. Amna Binte Kamran, Nigar Azhar Butt, and Amna Basharat. Semantic enrichment of Hadith corpus — knowledge graph generation from Islamic text. Semantic Web, 2026. doi: 10.1177/22104968261431425. ↗
  9. Sameh Mahmoud and Emad Nabil. Narrator identification by querying Sanad graph and utilizing the NarratorsKG on AR-Sanad 280K-v2 dataset. Neural Computing and Applications, 2024. doi: 10.1007/s00521-024-10194-2. ↗
  10. Sameh Mahmoud, Omar Saif, Emad Nabil, Mohammed Abdeen, Mustafa ElNainay, and Marwan Torki. AR-Sanad 280K: A novel 280K artificial sanads dataset for Hadith narrator disambiguation. Information, 13(2):55, 2022. doi: 10.3390/info13020055. ↗
  11. Sameh Mahmoud et al. Narrator2Vec: An efficient narrator representation in Hadith literature using word embedding. Arabian Journal for Science and Engineering, 2023. doi: 10.1007/s13369-023-08224-7. ↗
  12. Mohammed Mghari, Omar Bouras, and Abdelaaziz El Hibaoui. Sanadset 650K: Data on Hadith narrators. Data in Brief, 44:108540, 2022. doi: 10.1016/j.dib.2022.108540. ↗
  13. Abdelilah Mhamedi. Computational methodologies for sanad-based Hadith analysis: A review. TELKOMNIKA, 21, 2023. URL https://telkomnika.uad.ac.id/index. php/TELKOMNIKA/article/download/27447/12280. ↗
  14. Ahmad Mohasseb, Mohamed Bader-El-Den, and Mihaela Cocea. A hybrid algorithm for matching Arabic names, 2013. URL https://arxiv.org/pdf/1309.5657. ↗
  15. Multi-IsnadSet MIS Team. Multi-IsnadSet MIS for Sahih Muslim Hadith with chain of narrators, based on multiple ISNAD. Data in Brief, 2024. doi: 10.1016/j.dib.2024. 110284. ↗
  16. Muslimscholars. Muslimscholars.info: Database of Islamic scholars, 2026. URL https://muslimscholars.info. Accessed June 2026. ↗
  17. Maxim Romanov. Algorithmic analysis of medieval Arabic biographical collections. Speculum, 92(S1), 2017. doi: 10.1086/693970. ↗
  18. Maxim Romanov and Mamoona Seydi. Creating topological networks for Arabic biographical dictionaries. Technical report, HAL, 2022. URL https://hal.science/ hal-03533506v3/document. hal-03533506. ↗
  19. Sobia Saeed, Sobia Yousuf, Faisal Khan, and Qasim Rajput. Social network analysis of Hadith narrators. Journal of King Saud University – Computer and Information Sciences, 2021. doi: 10.1016/j.jksuci.2021.01.017. ↗
  20. Taufiq Wirahman. Cross-Source Hadith Narrator Entity Resolution: Linked biographical and transmission-graph data, 2026. URL https://doi.org/10.5281/zenodo. 21019693. Zenodo. ↗

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wirahman-2026-linking-hadith-narrator-identities-across/