AI Readers Advisory: Synthetic Reader Profiles (2023)
PRAIRIE (Public Readers Advisory Intelligent Recommendation Innovation Experiment :-)
At The Galecia Group, we know public libraries. So we also know that their staff wear many hats and do many important duties - but one of the most important roles is Readers Advisory.
The Internet definitely changed how Readers Advisory is performed. Websites and services channel the preferences of millions of readers and book purchasers into a river of data. But the Readers Advisory librarian is still critical: instead of helping patrons navigate the stacks in a single library building, RA librarians now help patrons navigate the millions of items and thousands of review and recommendation sources available online.
{{prAIrie}}: Profile-Based Recommendations
Does a complex reader profile yield better recommendation results?
This is a preliminary investigation. First, we generated 200 simulated reader profiles with detailed likes, dislikes, and search targets across a variety of fiction genres. Then we submitted each reader profile to GPT-3.5 with a two-part prompt as detailed below.
For each reader profile, we request five recommendations that meet the search target. Each recommendation contains title, author, and a rationale for why the reader might enjoy the recommended book.
Example Profile
| Attribute | Description |
|---|---|
| Name | Jasper McWhiskers |
| Favorite Genre | Science Fiction |
| Likes | Jasper enjoys reading science fiction novels that explore futuristic worlds and innovative technologies. He particularly enjoys books with hard science elements and realistic depictions of space travel and exploration. He also likes stories that delve into the psychological and philosophical implications of technological breakthroughs. |
| Dislikes | Jasper doesn't care for science fiction stories that are too focused on action and adventure, and he generally avoids space operas and military sci-fi. He also dislikes books with simplistic or cliched depictions of futuristic societies. |
| Favorite Authors | - Arthur C. Clarke - Kim Stanley Robinson - Neal Stephenson |
| Searching for | Jasper is in search of a science fiction novel that explores the potential consequences of artificial intelligence and machine learning. He is interested in a story with well-rounded characters and a plausible depiction of how AI might change our world. Jasper is also drawn to stories with an environmental message or other social themes. |
Example Results
| Title | Author | Rationale |
|---|---|---|
| The Lifecycle of Software Objects | Ted Chiang | Explores the relationship between humans and AI, with ethical implications of sentient beings and thought-provoking narrative. |
| The Water Will Come: Rising Seas, Sinking Cities, and the Remaking of the Civilized World | Jeff Goodell | Non-fiction on climate change impact and rising sea levels, touching on social justice and inequality. |
| The Diamond Age | Neal Stephenson | Explores nanotech and AI consequences, weaving education, culture, and social change. |
| The Windup Girl | Paolo Bacigalupi | Set in a world ravaged by crises, explores bioengineering, AI, and environmental consequences. |
| The Singularity Trap | David Dalglish | Fast-paced AI singularity novel exploring identity, morality, and consciousness. |
Results
We created 200 reader profiles. For each profile, we requested 5 title suggestions.
Our 1,000 suggestions included 334 unique titles from 283 unique authors.
First lesson learned: don't let ChatGPT pick the genres:
| Genre | Count |
|---|---|
| Science Fiction | 57 |
| Young Adult | 32 |
| Horror | 31 |
| Fantasy | 27 |
| Romance | 21 |
| Espionage | 13 |
| True Crime | 6 |
| Historical Fiction | 3 |
| Dystopian Fiction | 2 |
| Graphic Novels | 2 |
| Adventure | 2 |
| Biography | 1 |
| Humor | 1 |
| Comedy | 1 |
| Realistic Fiction | 1 |
Top 5 Recommended Titles
| Title | # Times Recommended |
|---|---|
| The Fisherman | 29 |
| The City We Became | 26 |
| The Fifth Season | 26 |
| The Diamond Age | 25 |
| The Left Hand of Darkness | 23 |
Top 5 Recommended Authors
| Author | # Times Recommended |
|---|---|
| N.K. Jemisin | 53 |
| Ursula K. Le Guin | 46 |
| John Langan | 29 |
| Neal Stephenson | 27 |
| Shirley Jackson | 21 |
Analysis
As noted earlier, we let ChatGPT pick the favorite genres of each simulated reader. As shown in the system prompt below, the one-shot training example was based on science fiction, which could explain the predominance of sci-fi in the results.
Even when we specified a wide list of potential genres in the prompt, the system gravitated toward SF, YA, and Horror.
Available genres suggested to the simulator: Fiction, Mystery, Science Fiction, Fantasy, Romance, Historical Fiction, Graphic Novels, Poetry, Drama, Short Stories, Adventure, Young Adult, Children's Literature, Picture Books, Realistic Fiction, Self-Help, Thriller, Horror, Biography, Autobiography, Paranormal Fiction, Dystopian Fiction, Crime, Western, Espionage, Horror, True Crime
When we conduct the full version of this experiment, we'll need to specify a balance of genres.