Data Skeptic

Book Ratings and Recommendations

39 min

Hannes Rosenborsch discusses his research on book ratings and recommender systems, primarily using Goodreads data. His work reveals that aggregated book ratings are poor predictors of individual enjoyment, as differences between books are minimal compared to the significant variance introduced by individual readers' rating habits. He advocates for content-based recommender systems, even building his own using LLMs, to help users understand their personal reading preferences rather than just predicting what they might like.

Summarized by Podsumo

🎧 Listen 🎙️ Ask about this episode

✨ Key Takeaways

1

Book quality is subjective: Research on Goodreads data shows minimal differences in ratings between professionally published books, while reader biases significantly impact scores.
2

Aggregated ratings are unreliable for individuals: Even with thousands of ratings, a book's average score (e.g., 4.1 vs 4.3) is not a strong predictor of a specific person's enjoyment.
3

Content-based recommenders are preferred: Hannes champions systems that analyze book attributes to help users understand *why* they like certain books, fostering self-insight over blind predictions.
4

LLMs as research assistants: Large Language Models are effectively used for annotating book features and even hypothesizing user preferences in Hannes's "Isaac method."
5

LLMs in creative writing: While useful for research and utility tasks (like finding words or brainstorming), LLMs are not used by Hannes for actual creative writing, as it defeats the purpose of self-expression.

💬 Notable Quotes

"The differences between books, especially professionally published books as you most encounter on Goodreads, are very minimal. Most books, especially if they have a couple of ratings, only they just fall into the same corridor of ratings and the differences between them are to use a loaded term, not significant or anything like that. But we see the opposite for readers."

— Hannes Rosenborsch
"I don't enjoy these collaborative filtering things, where it's just based on your history. Other people like this and then I don't understand anything about myself."

— Hannes Rosenborsch
"As a fiction author, you're expressing who you are, you're expressing what you care about, and you are so opinionated about every single word."

— Hannes Rosenborsch