You are here
Research: ChatGPT and Readers Advisory
Posted by Jim Craner on June 6, 2023
In our Library AI workshops, we mention the potential of AI-based tools like ChatGPT to assist with library services like Readers Advisory ("RA"). However, until recently, we only had anecdotal examples of how ChatGPT performed at this task.
So, we've spent some time lately figuring out how to methodically explore ChatGPT's capability as an RA tool. It's slow going -- we've just dipped our toe in the water, so to speak -- but we see some great potential!
We always have to include this disclaimer: we don't think that an AI tool can replace a trained and caring librarian in the library; but we do want to see if an AI tool can assist that librarian.
For now, we'll be compiling our AI/RA research on our new PRAIRIE website. (PRAIRIE = Public Readers Advisory Intelligent Recommendation Innovation Experiment because I can't resist an acronym, ever. :-)
At this point, we're really just doing feasibility testing: trying and determining the best ways to explore the use of ChatGPT as an RA tool. So far, we've done some small tests:
- Author Suggestions: we gave ChatGPT an author and asked for three similar authors we might enjoy. Then we repeated that over 1,500 times and analyzed the results.
- Title "Readalikes" Suggestions: we picked 3,000 books at random and asked ChatGPT to recommend readalikes and to justify its recommendations. We plan to set up a way for librarians to vote on the best recommendations and help us "judge" ChatGPT's performance.
- Simulated Reader Profiles: we had ChatGPT generate 200 simulated reader profiles, sophisticated sets of reader preferences and search goals, then requested recommendations based on each of those profiles. There's some promising data here!
We've got some exciting ideas for moving this research forward!
- Increase suggestion database to tens of thousands of books and authors
- Continue work on an RA chat interface for gathering reader preferences
- Researching the use of RA tools to integrate directly with ILS for availability checks and holds
- Setting up crowdsourcing tools we can judge ChatGPT's performance together.
What would you like to see us research? Have you worked on AI + RA recently? Reach out and let us know!