A Cleaned and Annotated Dataset of Rings from the Metropolitan Museum of Art Open Access Collection

This deposit is a cleaned, structured, and annotated dataset of 1,364 finger rings from the Metropolitan Museum of Art Open Access collection, together with the code that produces it. The Met's original catalog records are written for cataloging rather than analysis: materials are recorded as free text, culture and date labels are inconsistent, and ring-shaped objects that were never worn on a hand (ring bottles, sword pommels, harness rings) are filed under the same names as finger rings. This dataset resolves those problems for one object type. Each ring carries eleven fields: a stable Met object identifier, one of nine culture groups, modern country, a single representative year (negative for BCE), the primary material and the full list of materials, one of nine functions (ornamental, signet, archer's, scarab-ring, amuletic, mourning, devotional, betrothal, posy), the setting type, the iconography, whether the ring is inscribed, and a label_source flag. The visual fields (setting, iconography, inscription) are read from the object's image where one exists and inferred from the catalog text otherwise, as recorded by label_source. Every field uses a documented controlled vocabulary, defined in `data/vocabularies.md`. The rings span roughly 3335 BCE to 2005 CE and 37 modern countries across North Africa and the Nile Valley, the Mediterranean, Europe, the Near East, Central and East Asia, Island Southeast Asia, and the Americas. The dataset was derived from the full Met Open Access catalog (about 485,000 records): records were filtered to ring-named objects, the free-text fields were normalized, function and related attributes were labeled from object images and text, and ring-shaped objects that are not worn rings were removed. Normalization and labeling were carried out with a large language model (Gemini 3.1 Pro) under author-defined controlled vocabularies and a fixed schema, and the image-based function labels were checked against the Met curators' own labels, giving Cohen's kappa = 0.79 ("substantial" agreement). The data is published as a single UTF-8 CSV with a field dictionary, released under a CC0 1.0 public-domain dedication, and is ready for comparative and computational study of personal ornament across cultures and periods. The notebook in `code/` reproduces the dataset's overview figure from the CSV.