Creating a Space for Diversity and Inclusion with Artificial Intelligence
Formally located on the second floor of the Langson Library, the “Ask Us” Reference Desk was once the first stop for many students looking for book recommendations for leisure reading. This type of recommendation, commonly referred to as “readers’ advisory” is not typically a core service at academic research libraries like UCI Libraries. When the Libraries adopted OverDrive, a digital platform where users can borrow popular press eBooks, the number of “readers’ advisory” requests increased exponentially. Library Assistant, Kelsey Brown, often found herself offering general book recommendations to the Libraries’ users. Inspired by her conversations with patrons, Brown wondered if it would be possible to train ANTswers, the UCI Libraries’ experimental chatbot, to provide personalized book recommendations to patrons. Through conversations with Computational Research Librarian and creator of ANTswers, Danielle Kane, it was determined that ANTswers could be coded to provide these recommendations.
As she researched book titles to add to the enhanced code, Brown noticed an interest among readers in books by authors of color on a variety of topics. “I felt it was important to increase the number of titles directly related to racism and prejudice and have ANTswers recommend a wide variety of genres by authors of color,” shares Brown. She compiled a list of 729 book titles, with over 30 “genre” tags from broad “love story” or “romance” to the more specific “domestic fiction” and “movie adaptation.”
Brown imagined two scenarios in which ANTswers provide recommendations:
- A patron named a book they liked (e.g. the most popular fiction book borrowed on Overdrive, Little Fires Everywhere) and ANTswers would recommend a similar book.
- The patron would type in a genre and ANTswers would recommend a book.
In both scenarios, ANTswers would only name books available through UCI Libraries. To reduce choice paralysis for patrons, Kane wrote code that recommends a single book per search and a different book each time the same request is typed in.
Kane enlisted the help of San Jose State University Master’s of Library and Information Science candidate and the Libraries’ student intern, Danielle Dantema, who created some of the initial files for a few new keywords. “There is no room for mistakes when writing code — if there is one comma, letter, semicolon, period, or any necessary character out of place or missing then the whole bot will not work,” shares Dantema. Another issue to resolve before launching the new system was ensuring newly developed code would not conflict with previous code that ran queries in Library Search.
In addition to the meticulousness of writing code, the project provided an opportunity to learn more about the thought processes behind retrieving a book title. Chat Bots must be programmed to recognize phrases people use and human errors, such as grammar and misspellings, when searching for a title, genre, or topic.
ANTswers is one of few academic library chatbots in the country and one of the longest running. “With this homegrown system there are no barriers to access. Questions about the Libraries are answered 24/7 with minimal downtime and adding the new module of book recommendations was a natural extension of what ANTswers does,” shares Kane who is proud of the latest enhancements.