CoolFace
Datasetpublic

FatimahEmadEldin/Saudi-historical-places-v2

Saudi Historical Places v2 (verified) A multimodal Arabic corpus of Saudi historical sites, spanning all thirteen administrative regions. Each record pairs curated structured attributes with encyclopedic narrative text and screened photography. This is a corrected release. Version 1 linked each site to an Arabic Wikipedia article using title lookup with a fuzzy fallback; auditing that linkage found that 31.4% of links (Wilson 95% CI 26.8-36.4, n=357) denoted a different place… See the full description on the dataset page: https://huggingface.co/datasets/FatimahEmadEldin/Saudi-historical-places-v2.

sourceHugging Facecc-by-sa-4.0updated 2mo agoView on Hugging Face
0likes41downloads
Dataset Card

Saudi Historical Places v2 (verified)

A multimodal Arabic corpus of Saudi historical sites, spanning all thirteen administrative regions. Each record pairs curated structured attributes with encyclopedic narrative text and screened photography.

This is a corrected release. Version 1 linked each site to an Arabic Wikipedia article using title lookup with a fuzzy fallback; auditing that linkage found that 31.4% of links (Wilson 95% CI 26.8-36.4, n=357) denoted a different place, with substantial inter-adjudicator reliability (Cohen's kappa = 0.835).

What changed from v1

ChangeEffect
Region taxonomy reconciled16 raw values -> 13 official regions (region_canonical)
Dates parsedHijri/Gregorian split, centuries and named eras recognised, 289 years recovered by cross-calendar inference
Links verifiedEvery narrative link audited; unverifiable links withdrawn rather than kept
Links repairedCorrect articles found where possible; coverage extended to sites that had none
Imagery re-collectedPhotographs screened for whether they actually depict the site
Captions addedArabic and English captions per image

Trustworthy narrative coverage rose from 222 to 371 sites (+67%).

Evidence policy

No article is better than a wrong article. A mislinked narrative does not merely fail to help a retrieval system; it supplies fluent, confident, irrelevant prose that retrieval ranks and generation cites. Where no link passed verification, the narrative field is NULL and link_status records why. The curated الوصف summary is present for every record, so every site remains answerable regardless.

Key columns

  • —region_canonical, region_raw, region_method — normalized region plus how it was resolved
  • —hijri_year, gregorian_year, era_id, date_certainty — parsed dates; NULL means unknown, never imputed
  • —link_status — verified_original / verified_relinked / verified_kept / no_verified_article
  • —link_verdict_original — the audit verdict on the v1 link (verified / suspect / mismatch)
  • —caption_ar, caption_en — vision-language captions of the site photograph
  • —source_branch — provenance: curated or places_branch_only

Known limitations

  • —Coverage is geographically uneven; المدينة المنورة is 36.7% of records.
  • —Narrative text inherits Arabic Wikipedia's own coverage and editorial biases.
  • —Calendar conversion is year-granular (±1 year); century and era values are midpoints for ordering only, not date estimates.
  • —73 sites have no photograph that was confirmed to depict them.
  • —Photographs derive from a commercial mapping API; redistribution terms apply.

Narrative text is Wikipedia-derived and therefore CC-BY-SA; attribution and share-alike propagate.