Piloting Wikidata as an Authority Identifier: The (In)visible Women Project at the Smithsonian Institution

Phase One: May 2022–April 2023

Units across the Smithsonian use different collections information systems, including EMu (Electronic Museum), TMS (The Museum System), and ArchivesSpace. These systems do not speak to each other and units prioritize different data points to include within their people records, as reflected in their collecting policies and scopes. These workflows often result in duplicate people records across the Smithsonian.

Phase One of the project focused on data cleanup. A contractor disambiguated names provided by the 10 participating Smithsonian units and associated existing identifiers from external sources to people records. Each participating Smithsonian unit sent datasets of person names and relevant biographical information. Using OpenRefine, identifiers for each person name were located and imported from several databases: Social Networks and Archival Context (SNAC), Union List of Artist Names (ULAN), Virtual International Authority Files (VIAF), and Wikidata. The Library of Congress Authority Files (LCNAF) identifiers were also imported manually using links from other identifiers or by searching in the Library of Congress Name Authorities. OpenRefine software was used to access identifiers via APIs in order to automate this part of the process; even so, each name required research and disambiguation to find the appropriate identifiers. Each relevant unit received a spreadsheet of the gathered identifiers for quality checks.

Of the over 20,000 people records reviewed by the contractor, 8,700 had existing external identifiers. However, it was determined that over 11,600 names had been locally created (see Table 1).

Phase Two: May 2024–August 2025

During this portion of the project, we created 747 Wikidata records, drawing on the local names specifically. We also edited 720 existing people records in Wikidata, linking them back to Smithsonian collections that they are affiliated with. For all of the records created and edited, we have linked them to the American Women’s History Initiative Focus List on Wikidata.

The information that we entered into these records is based on a data model we developed collaboratively with the participating units. At the start of phase two, we conducted informational interviews with the project partners using semi-structured interview methods (Lui & Wildemuth, 2009). During the interviews, we asked questions about data management within the interviewees’ units, their database workflows, what types of information they find important to capture about people represented in their collections, and what names they see as a priority for Wikidata record development. Based on our preliminary work with Wikidata and insights gleaned via thematic coding (Saldana, 2016), we developed a flexible data model available on our Wikiproject page. Every field is optional to account for varying degrees of data maintained across Smithsonian units.

With the support of Smithsonian colleagues, we ingested all external identifiers gathered and created through phase one and two into multiple Smithsonian databases, including ArchivesSpace, TMS, and EMu. We have also collaborated with participating unit representatives and with the Office of Digital and Innovation to build shared practices on where to document external authority identifiers in the various databases used at the Smithsonian. For example, in ArchivesSpace, identifiers are ingested into the record IDs section, and in TMS, identifiers are ingested as URLs in the alternate numbers field.

Our work in Wikidata, as a public and collaborative platform, prompted consideration about when we should not make a Wikidata record for a given person. Based on conversations with the Smithsonian Privacy Office, we developed a decision tree that considers a number of legal and privacy questions (see Figure 1). For example, in Smithsonian records, is this person a named donor, archival creator, or artist/maker? If so, Smithsonian staff are in a position to share information about that person (Smithsonian Institution, 2024). For this document, we have drawn on research concerning legal definitions of who counts as a public figure (Legal Information Institute, 2020), scoping outlines from ULAN and LCNAF (Library of Congress, n.d.; Getty Research Institute, 2024), and notability guidance from Wikipedia (Wikipedia Notability, n.d.; Wikipedia Notability (people), n.d.; Wikipedia Notability (academics), n.d.).

Figure 1

Data Model and Dataset Limitations

Even though this project aimed to remedy persistent data gaps, the resulting dataset has limitations, per the institutions and data models that shaped it. The names in this dataset were composed by the collecting policies of the partnering Smithsonian units, each with their own histories and focuses. Further, the databases in use at each Smithsonian unit have different search capacities. Some of the database exports of people names were scoped by a recorded gender identity and others were not.

We also encountered affordance challenges from using Wikidata as an authority identifier source. For example, Wikidata lacks accurate statement options for documenting Tribal affiliation. We considered a number of potential statements, such as “country of citizenship” and “ethnic group,” but these options felt either inappropriate for our use as digital curators or inaccurate. We eventually settled on including Tribal affiliation in the description field as it is searchable there. This impacts possible SPARQL queries, however, as this search method cannot be scoped by Tribal affiliation as a result of our data model decision. Additionally, our data model does not include a “sex or gender” statement, as in the majority of cases we did not know gender identity from the perspective of the individual reflected in a given record (see The Trans Metadata Collective et al., 2025).