What Your Favorite Books Reveal About Personality

Favorite Books and Personality: What’s the Connection?

Do people who love fantasy novels have different personalities from people who devour self-help books? It’s a question worth asking if you’re building a book recommendation service, because what you can learn about a reader from the books they like shapes what you recommend next.

A research team led by Ng Annalyn analyzed data linking personality surveys from over 60,000 Facebook users to the “likes” they’d left on book pages. When they grouped personality scores by readers who liked the same book, those scores tracked with the tags attached to each book. What matters here isn’t how accurately the model predicted any single individual, but the differences that showed up between reader groups for different books.

The tags weren’t limited to genres like fantasy or history — they also included reader-generated labels like “to read later.” The research team argued that this kind of granular detail can surface distinctions that broad genre categories tend to miss.

Two Ways to Measure Personality: MBTI and the Big Five

This study measured personality using the Big Five (OCEAN)¹ — a framework that scores five traits individually: openness, conscientiousness, extraversion, agreeableness, and neuroticism.

In Korea, MBTI comes up constantly in conversation. Where MBTI sorts people into 16 types, the Big Five treats each trait as a continuous score showing how much of it someone has. The exact range of the scores depends on the survey instrument and scoring method used.

For example, under the Big Five, someone can score high on extraversion and high on openness at the same time. It’s the combination of all five scores that captures differences between people. So you can’t meaningfully compare the two frameworks by tallying categories — asking whether 16 types is “more granular” than 5 traits misses the point entirely.

What matters for this piece isn’t which test is superior, but what the scores actually mean. A personality trait score is not a report card grading someone’s character or ability. And no one should conclude that reading a particular genre makes a person conscientious — or the opposite.

The draft looks accurate and complete against the source. No corrections needed.

We Grouped 60,000 Responses by Book

The research team analyzed 479 Facebook book pages. Each page had been liked by at least 50 users who had completed the personality survey, and the total pool, once duplicates were removed, came to 61,662 users. The team used the median personality score of the users who liked each book as that book’s representative score. Because this is based on “likes,” the data doesn’t confirm whether people actually read the books.

Book content was captured through tags from the reading community Goodreads. The dataset they compiled — which included books beyond those analyzed — covered 14,731 books, 24,091 distinct tags, and roughly 193,500,000 tag-usage records. The team consolidated tags with similar meanings into 396 groups. They then used each book’s tag composition to predict the personality scores of its readership.

Which Tags Correlated With Which Personality Scores

Here’s what the paper reported as correlations. These are all tendencies of reader groups aggregated by book — they don’t apply directly to any individual.

Extraversion: Tags like relationships, chick lit,² and memoir were associated with higher extraversion scores. Fantasy, sci-fi, and comics were associated with lower scores.

Agreeableness: Tags for books dealing with family or religion were associated with higher agreeableness scores, while tags like psychological drama or cult classic were associated with lower scores. There’s no basis for interpreting a work’s tone as a reflection of its readers’ morality.

Openness: Tags related to philosophy or classic literature were associated with higher scores, while light fantasy was associated with lower scores. In a random forest model that combined multiple tags, the R² for openness was 0.47.³ This means the model explained about 47% of the variation in median openness scores across reader groups by book. It should not be read as meaning the model guessed any one person’s personality with 47% accuracy.

Neuroticism: This trait refers to a tendency to experience anxiety or negative emotions. Tags for sad endings and mental health were associated with higher scores. As for the correlation with the “pretty cover” tag, the research team itself noted that gender differences might have played a role.

Conscientiousness: Tags tied to practical content — professional knowledge, business, and the like — were associated with higher scores. Humor and young-adult books were associated with lower scores, though the team also floated age differences among readers as a possible explanation.

The “Books for Later” Tag Was Part of the Analysis, Too

“Back-burner” is a tag marking books a reader has set aside to read later. This tag group’s correlation with openness was 0.28 — the highest positive correlation between any tag and personality trait reported in the study.

Here again, it’s worth remembering where the data came from. The people who applied the tags were Goodreads users; the people who took the personality survey were Facebook users. The two groups were linked only through the books themselves, so this doesn’t confirm that respondents high in openness personally piled up unread books. Still, it does show that markers readers themselves attach to books — beyond the content itself — can also be used in this kind of analysis.

How Could This Be Used for Recommendation Services

Classifying content in finer detail gives recommendation systems more to work with. Netflix is a case in point. In 2014, The Atlantic identified and analyzed 76,897 micro-genres on Netflix. These combined broad categories like romance or action with era, setting, mood, and other attributes. That figure reflects the state of the survey at the time.

What the book study and the Netflix case have in common is that they describe content in far more granular terms than a broad genre label allows. Pairing this kind of classification with a user’s preference history can help surface titles they’re likely to enjoy.

That said, fine-grained genre classification is not itself a personality test. The book study presents no evidence that Netflix’s recommendation system measures the Big Five. Predicting what someone will watch next and predicting their personality-survey scores are, from the outset, different goals.

There is separate research that predicts individual personality scores from online preference records. In 2015, a Cambridge University team led by Wu Youyou analyzed personality surveys and Facebook Like data from 86,220 people. In the team’s comparisons, the model surpassed the judgment accuracy of friends using about 70 Likes, family using 150, and spouses using 300. The benchmark was how closely the model’s predictions matched the person’s own self-reported survey answers. This doesn’t mean that a handful of Likes reveals anyone’s inner life.

In Korea, too, more people are leaving digital traces of their reading habits. In the “2025 National Reading Survey,” released in March 2026 by the Ministry of Culture, Sports and Tourism, the overall reading rate among adults was 38.5%. Among 19-to-29-year-olds surveyed, the rate was 75.3%, up 0.8 percentage points from 2023. Within this age group, the e-book reading rate was 59.4% and the print-book reading rate was 45.1%. The overall reading rate refers to the share of people who read or listened to at least one general book in the past year.

Depending on their features and data-collection policies, e-book services can log which books someone has read, where they left off, and how long they spent reading. More records mean more raw material for recommendations, but how useful that data actually is for predicting personality is a separate question that requires its own verification. This survey alone says nothing about whether — or how accurately — any platform analyzes personality.

Oswarld’s Lens

As someone who’s worked in data analysis, I found it interesting that they used tags people attached themselves as the basis for analysis. You can see more granular differences that way than if you just sorted books into a single genre. That said, when applying results like this to a business, you have to start by checking the unit of analysis. If a model describes clusters of books grouped together, using that same model for individual recommendations requires re-validating how well it actually performs against real users.

This also brought to mind things I’ve seen while building business strategy. When companies aim to understand users better, that goal is often bundled together with raising conversion rates and reducing churn. Netflix’s own researchers, in a 2015 paper, explained that they evaluate recommendation algorithms by looking at member retention and engagement. We need to look at both what counts as a good recommendation for the reader and what metric the company is actually trying to move.

What concerns me is that it’s hard for users to know just how far their data trail gets used. A feature that remembers where you stopped reading, one that recommends books, and one that infers your personality — users may have very different expectations about the scope of each. I think services need to clearly explain what information they collect and for what purpose.

Personally, I think there’s an opportunity here for book-club services like Trevari or Hitch too. Based on reading preferences and group preferences that members consent to share, these services could recommend the next book or the next group. I haven’t actually confirmed what tags or personality data these particular services hold. But as a business idea, I’d want to first verify whether you can build recommendations that members are actually satisfied with.

Closing

The books you love hold clues to who you are. This study looked at those clues through book tags and the personality scores of reader groups. But liking a single book isn’t enough to pin down someone’s personality — that would go beyond what the research actually shows.

So does reading something different from your usual fare change your personality? In a 2009 experiment by Djikic et al., 166 participants were split into two groups: one read a short story by Chekhov, the other read a factual account of the same events. The group that read the fiction showed a larger shift in self-reported personality traits immediately afterward. The experiment didn’t test whether this change lasts, or whether reading a specific genre can make you into the person you want to be.

Rather than setting out to change my personality, I’d rather just pick up a book from a genre I don’t normally read. After all, you might only discover something new to love by reading a book that was never on your recommended list.

References & Further Reading

• Ng Annalyn et al., Predicting Personality from Book Preferences with User-Generated Content Labels, preprint released 2017, published in IEEE Transactions on Affective Computing, 2020. The book-level aggregation method appears in Section 2.1, tag cleanup in Section 3, and prediction results in Section 4.1.
• Wu Youyou et al., Computer-based personality judgments are more accurate than those made by humans, PNAS, 2015. University of Cambridge research summary.
• Ministry of Culture, Sports and Tourism (Korea), 2025 National Reading Survey Results, March 6, 2026. Note that the year in the survey’s title differs from the year it was published.
• Alexis C. Madrigal, How Netflix Reverse Engineered Hollywood, The Atlantic, January 2, 2014.
• Carlos A. Gomez-Uribe and Neil Hunt, The Netflix Recommender System: Algorithms, Business Value, and Innovation, 2015. This explains the evaluation goals behind Netflix’s recommender system.
• Maja Djikic et al., On Being Moved by Art: How Reading Fiction Transforms the Self, Creativity Research Journal, 2009.

The author is Oswarld (Kwangseob Ahn). Current roles: Adjunct Professor at Sejong University, Strategy Consultant at INLEVEL9. Career, research, books, and recent work are kept current on the About page. Latest · July 2026: HEMA-2: A Consolidation-Aware Tri-Memory Architecture with Multi-Channel Scheduling for Lifelong Conversational AI.

Footnotes

1. Big Five (OCEAN): a model that describes personality through five traits — Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism. Each trait is measured on a continuous scale, and the exact scoring range varies by assessment tool. ↩

2. Chick Lit: a genre label used for novels that treat the daily life, romance, and work of mostly young women with humor. “Bridget Jones’s Diary” is a classic example. ↩

3. R² (coefficient of determination): a measure of how much of the variation in the target values a model explains. The 0.47 cited in this paper comes from a random-forest model using multiple tags to predict Openness. The target being predicted is the median score of a book’s reader population, not any individual’s score, so this differs from person-level accuracy — and the number alone can’t settle how useful the model is in practice. ↩