book vibes

This tool uses book embeddings learned from the goodbooks-10k dataset. The embeddings were generated in Python using rating collocation and hybrid rating/tag methods — the code for that lives at ryebreado/book-vibes.

When you select seed books, the tool computes the centroid of their embedding vectors and ranks every other book by cosine similarity to that centroid. A popularity penalty can optionally down-weight mainstream titles that tend to dominate any literary centroid:

raw score = cos(centroid, book) − w · log(1 + popularity)

Scores are then scaled 0–100 relative to the top result. The raw cosine similarity is shown alongside for reference.