April 2, 2020
Search engines are a ubiquitous part of our lives in this age of information, but we rarely stop to think about how all the information at our fingertips made it into our search results. This concept is at the heart of work performed by the Digital team at The Frick Collection. Our online collection—which allows the public to search through the works of art held by the Frick, including those not on view, and even organize them into personalized groupings—depends on reliable data. When someone enters a query into a search box, they often assume the results are accurate. It’s up to those of us tasked with managing stores of information to make sure that’s true.
The Challenge
In the past, if a visitor to the Frick’s collection website searched the term “portrait,” the search results did not display all the portraits from the permanent collection, unless that term was in the title of the work. In order to rectify this, the first order of business was to add broad keywords such as portraits, landscapes, and cityscapes to objects in our collections database.
Next, we had to determine how to narrow that down and build a richer database of keywords and subjects. What terms are website visitors most interested in searching, and how do you account for variations of a term? Does every subject in a painting need to be described, like a small bird made with two brushstrokes in the sky of a landscape painting? And, finally, what’s the most efficient way to assess every object in the collection?
Research
Research on this project included reviewing the website’s Google Analytics data to see what terms visitors were entering when using the site’s built-in search function. The most popular artists in the collection were at the top of the list—Vermeer, Ingres, Rembrandt—but so were terms like "portrait," "landscape," and "love." The Frick’s curators also helped to determine what terms were most important, drawn from the Getty Art and Architecture Thesaurus (AAT) and based on their own knowledge of the collection. While aiding researchers at the Frick Art Reference Library, the Assistant Librarian for Public Services identified further terms when visitors' searches on the website didn’t produce sufficient examples from the collection. With this cross-departmental support we were able to get an idea of what visitors were most interested in searching.
Further investigation for the project included learning about how other institutions have increased access to their collections and the potential use of artificial intelligence (AI). Notable examples include SFMOMA and The Metropolitan Museum of Art. At the Frick, we also experimented with computer vision AI, running tests with the Google Cloud Vision API. This digital image analysis tool was able to detect very obvious keywords, but its accuracy was questionable, especially with emotionally laden terms like “love.” In the end, we decided to stick with human tagging.
Tools and Technology
The project was accomplished using the Frick’s internal collections database, The Museum System (TMS). One way in which TMS was particularly useful for this project was with the creation of object packages. Object packages make it easy to break up thousands of records into more manageable chunks.
The next piece of technology that’s vital to the Digital team is eMuseum, a web publishing platform that integrates with TMS. This software pulls information about the collection and associated media onto the website. The content in TMS is published via eMuseum at collections.frick.org, and makes the collection available to all our website visitors.
Anyone can search our collection, now using keywords, and review premade collections of objects grouped by subject matter through the My Collections feature of eMuseum. You can even assemble your own groups of works from the Frick through My Collections to explore specific themes independently, whether for fun, for research, or for the classroom. Check out this blog post that walks you through the process.
What's Next
This project is ongoing. As we continue to assess popular searches and receive feedback from visitors and internal staff, new keywords will continue to be added across the collection. If you have searched our collection online, please share your keyword searches with us in the comments below so we can continue to improve.
Screenshot of a search for "portrait" at collections.frick.org.
Visual representation of keyword subject analysis, using a detail of Corot's The Lake.
Screenshot of My Collections on The Frick Collection website.